How to Run an ELISA: A Step-by-Step Protocol

Running an ELISA involves coating a microplate with your target molecule or capture antibody, blocking empty binding sites, adding your sample and detection reagents in sequence, and reading the resulting color change on a plate reader. The basic workflow has not changed dramatically since the method was first developed in 1971, when researchers replaced radioactive labels with enzyme conjugates to quantify immunoglobulin levels in rabbit serum.¹ What has changed is the range of formats, substrates, and detection strategies you can choose from, and the details at each step make the difference between clean data and a plate full of noise.

Choosing Your ELISA Format

Before you coat a single well, you need to decide which ELISA format fits your experiment. The four main types are direct, indirect, sandwich, and competitive, and each varies in how many antibodies you use, how the signal is generated, and what kind of question the assay can answer.²

Direct ELISA: You coat the plate with your antigen, then add an enzyme-conjugated primary antibody that binds to it. Fewest steps, but you need a labeled primary antibody for every target, which limits flexibility.
Indirect ELISA: Same antigen-coated plate, but you add an unlabeled primary antibody first, then an enzyme-linked secondary antibody that recognizes the primary. More flexible because one labeled secondary can work across many primaries.
Sandwich ELISA: You coat the plate with a capture antibody instead of the antigen. Your sample antigen binds to the capture, and then a detection antibody binds a different site on the antigen, followed by an enzyme-linked secondary. This is the workhorse format for quantifying proteins in complex samples like serum or cell culture supernatant.
Competitive ELISA: Sample antigen competes with plate-coated antigen for binding to a limited amount of primary antibody. The more analyte in your sample, the less antibody binds the plate, so the signal goes down as concentration goes up. Useful when your target is small or has only one antibody-binding site.

Sandwich assays are the most common choice for quantitative work because they use two antibodies recognizing different parts of the same molecule, which boosts specificity. Competitive formats tend to shine for small molecules like drugs or hormones where finding two non-overlapping antibody-binding sites is difficult. If you are simply screening whether a sample contains antibodies against a known antigen, indirect ELISA is usually the simplest route.

Step One: Coating the Plate

Coating is the step that anchors everything else. You dilute your antigen or capture antibody in a coating buffer and pipette it into the wells of a 96-well polystyrene microplate, then incubate overnight at 4°C or for a shorter period at 37°C to allow passive adsorption to the plastic surface. Carbonate-bicarbonate buffer at pH 9.6 is the most commonly used coating buffer, but it is not universally ideal. Some antigens lose activity in it. One study found that nervous necrosis virus antigens immobilized using carbonate buffer at pH 9.6 had dramatically lower ELISA signals compared with dry immobilization, and the loss of antigenicity was caused specifically by the carbonate and bicarbonate ions rather than the alkaline pH.³ If your antigen is pH-sensitive or has a viral coat structure, you may get better results with phosphate-buffered saline or a Tris-based buffer. Testing multiple coating conditions early on is well worth the plates it costs you.

Concentration matters too. For capture antibodies in sandwich ELISAs, something in the range of 1 to 10 micrograms per milliliter is typical, though the optimal concentration depends on your antibody’s affinity. Too little and you lose sensitivity; too much and you can actually reduce signal by crowding antibodies so tightly they sterically block each other. A checkerboard titration, where you vary both the capture antibody and detection antibody concentrations across rows and columns of the same plate, is the standard way to find the sweet spot.

Step Two: Blocking

After coating, plenty of bare polystyrene remains exposed. If you skip straight to adding your sample, proteins from the sample will stick to those empty spots and generate background signal that drowns out your real data. Blocking fills the gaps with an inert protein or polymer so that only specific antibody-antigen interactions drive the readout.

Bovine serum albumin at around 1 to 3 percent is the default blocking agent in most labs, but not all BSA preparations behave the same way. Researchers testing different Sigma BSA catalog numbers found that several gave high nonspecific binding, while only a preparation that had been depleted of both globulins and endotoxins eliminated the problem.⁴ If you are getting unexplained background and have already optimized your antibody concentrations, switching BSA lots or grades is one of the first things to try.

BSA is not your only option. Non-fat dry milk is cheap and effective for many applications, though it can interfere with biotin-streptavidin detection systems and assays targeting phosphorylated proteins. Synthetic blockers like Ficoll and polyvinyl alcohol perform comparably to BSA in buffer solution and are useful alternatives when you are detecting food allergens or other targets where animal-derived proteins could cross-react.⁵ Blocking typically takes one to two hours at room temperature, though some protocols use 37°C for speed or overnight at 4°C for convenience.

Step Three: Adding Samples and Primary Antibody

After blocking, you wash the plate several times with a buffer containing a mild detergent, usually PBS with 0.05 percent Tween-20. Washing removes unbound blocker and prepares the surface for your sample. In sandwich ELISAs, you add diluted sample and standards to the wells at this point, incubate for one to two hours, then wash again before adding the detection antibody. In indirect or direct formats, the antigen is already on the plate from the coating step, so you add the primary antibody directly after the post-block wash.

Standards are critical for quantitative work. You prepare a serial dilution of a known concentration of your target protein, which creates the standard curve your unknown sample concentrations will be read from. Most protocols use a seven- or eight-point curve plus a zero-concentration blank. Pipetting accuracy at this stage has an outsized effect on your final data. Even small volume errors in the serial dilution propagate through every downstream concentration point.

Incubation conditions should stay consistent. If one plate sits at room temperature for 90 minutes and the next sits for 120 minutes, the results are not directly comparable. Using a plate shaker during incubation improves uniformity by preventing depletion zones from forming at the bottom of the well, where bound antibody would otherwise consume local analyte faster than diffusion can replace it.

Step Four: Detection Reagents and Substrate

Once you have washed away unbound sample and detection antibody, you add the enzyme-linked component. In most setups, this means an enzyme-conjugated secondary antibody that recognizes the detection antibody. The two workhorses are horseradish peroxidase and alkaline phosphatase. A systematic comparison found little practical difference between them in terms of detection performance for the same analytes, meaning the choice often comes down to availability and what substrates you already have on hand.⁶ That said, HRP is by far the more common choice in colorimetric ELISAs because it is small, inexpensive, and works with fast-developing substrates.

For HRP-based colorimetric detection, TMB is the most widely used substrate. The enzyme catalyzes the oxidation of TMB by hydrogen peroxide, producing a blue color that turns yellow when you add an acid stop solution. TMB generates a visible, stable stopped signal, which makes it practical for standard plate-reader workflows.⁷ ABTS is an older alternative substrate, but its stopped signal is less stable on certain surfaces. For alkaline phosphatase, pNPP (p-nitrophenyl phosphate) is the standard chromogenic substrate, producing a yellow color read at 405 nm.

Timing the substrate reaction matters. You typically develop the color for 10 to 30 minutes in the dark, then stop the reaction with sulfuric acid for TMB or sodium hydroxide for pNPP. If you wait too long, the high-concentration standards saturate and the curve flattens at the top, compressing the dynamic range. If you stop too early, your low-concentration standards sit barely above blank and you lose sensitivity at the bottom of the curve. Watching the color develop and stopping when the highest standard is a deep blue but has not yet begun to plateau is the practical approach most experienced users rely on.

Step Five: Reading and Analyzing Results

You read the plate on a spectrophotometer, usually at 450 nm for TMB-based assays with a reference wavelength at 620 or 655 nm to correct for optical imperfections in the plate.⁸ The absorbance values from your standards generate the standard curve, and your software interpolates the unknown sample concentrations from that curve.

Standard curves in ELISAs are not linear. They are sigmoidal, with a flat bottom at low concentrations, a steep middle, and a flat top where signal saturates. The four-parameter logistic model is the most common fitting method, capturing the upper and lower asymptotes, the inflection point, and the slope. A five-parameter logistic model adds an asymmetry factor and can be useful when the curve is not perfectly symmetrical around the inflection point. In practice, the efficiency loss from using the five-parameter model when the curve is actually symmetric is negligible, so some labs default to it.⁹ The important thing is that your unknown samples fall within the steep, linear portion of the curve. Any sample reading near the upper or lower plateau should be re-run at a different dilution.

Sample Preparation Decisions That Affect Your Data

The choice between serum, heparin plasma, EDTA plasma, and citrate plasma is not neutral. Different anticoagulants can shift measured concentrations for many analytes. In one study comparing multiple sample types, the majority of cytokines and soluble proteins showed significant differences in mean concentration depending on the collection medium.¹⁰ For example, corin levels measured in EDTA or citrate plasma were significantly lower than levels measured in serum or heparin plasma from the same individuals.¹¹

Storage conditions matter just as much. Some analytes remain stable at room temperature or refrigerated for weeks in certain sample types but show drift in others. For tumor markers like CEA and PSA, only EDTA plasma maintained clinically acceptable bias across room temperature and refrigerated storage, while frozen storage introduced positive bias over time.¹² The practical takeaway is to decide on your sample type early, validate that your kit performs well with it, and keep the type consistent across all samples in a study. Mixing serum samples with EDTA plasma samples in the same dataset can introduce systematic bias that no amount of downstream statistics will fix.

Freeze-thaw cycles are another common source of trouble. Each cycle can denature proteins and reduce measurable concentrations. Aliquoting samples into single-use volumes before freezing avoids this. If you are working with archived samples that have already been thawed once or twice, note it and interpret your data accordingly.

Troubleshooting Common Problems

High background that obscures your signal can come from several places: inadequate blocking, insufficient washing, antibody concentrations that are too high, or contaminated reagents. The BSA purity issue described earlier is a real-world example. Work backward through the protocol to isolate the source. Running a plate with no sample but all other reagents tells you whether the background comes from your detection system or from sample matrix effects.

The edge effect is a well-known annoyance in microplate assays. Wells along the outer rows and columns tend to give higher or lower readings than interior wells. This happens because evaporation rates are higher at the plate edges, which concentrates the contents in those wells or alters temperatures. One modeling study found that conduction through the side walls of the plate caused lower liquid temperatures in edge wells, and that plates with transparent rather than opaque walls were inherently more resistant to the problem because air between the liquid and the wall acts as an insulator.¹³ A simpler workaround is to fill the outer wells with buffer only and use them as blanks, sacrificing some throughput for more uniform data in the remaining wells. Sealing plates during incubation also reduces evaporation.

The hook effect is a more dangerous artifact because it gives you falsely low results without an obvious red flag. In sandwich assays, when the analyte concentration is extremely high, excess analyte saturates both the capture and detection antibodies independently, so fewer sandwich complexes form and the signal paradoxically drops.¹⁴ The result looks like a low-positive or even negative sample when the true concentration is sky-high. Running samples at multiple dilutions catches this. If the undiluted sample gives a lower reading than the 1:10 dilution, you are likely seeing the hook effect and need to dilute further.

When Automation Makes Sense

Manual ELISA is perfectly adequate for small experiments, but as soon as you are running dozens of plates per week, human error and fatigue start to erode data quality. Semi-automated and fully automated systems handle liquid dispensing, washing, incubation timing, and plate reading with better precision than a tired pair of hands at 4 PM on a Friday. A comparison of manual versus semi-automated processing of a SARS-CoV-2 nucleocapsid IgG ELISA found that analytical performance was preserved after the transition, with precision staying below 25 percent relative standard deviation and accuracy below 20 percent error for all samples.¹⁵

Beyond matching manual performance, automated platforms bring advantages that are harder to quantify on a single plate but compound across a project. They can pipette volumes as low as one microliter accurately, run multiple assays simultaneously, and free up skilled technicians for work that actually requires judgment.¹⁶ The most meaningful benefit for many labs is eliminating the pipette-volume variation that creeps in during long manual runs. If you are performing a clinical study where regulatory agencies will scrutinize your assay’s coefficient of variation, automation is less a luxury than a practical requirement.

Chemiluminescent and Ultrasensitive Alternatives

Standard colorimetric ELISA works well for most applications, but when you need to detect very low concentrations, swapping the chromogenic substrate for a chemiluminescent one can push the detection limit down substantially. A comparison of colorimetric and chemiluminescent ELISA for anti-erythropoietin antibodies found that the chemiluminescent version reached a detection limit of 0.08 ng/mL versus 0.6 ng/mL for the colorimetric method, and the measurement range expanded more than twenty-fold.¹⁷ In food safety testing, a chemiluminescent ELISA for the pesticide imidacloprid achieved a detection limit of 0.19 micrograms per liter, compared with 1.56 micrograms per liter for the colorimetric version.¹⁸

At the high end of sensitivity, single-molecule array technology can detect biomarkers in serum at femtomolar concentrations, far below what even chemiluminescent ELISA achieves. A comparison of single-molecule array, high-sensitivity ELISA, and fully automated chemiluminescent immunoassay platforms for measuring neurofilament light chain in multiple sclerosis patients found strong correlations between single-molecule array and high-sensitivity ELISA, with no systematic bias between them. The fully automated chemiluminescent platform, however, showed higher measured levels when using frozen samples and was more affected by pre-analytical handling.¹⁹ The lesson for anyone considering these platforms is that increased sensitivity does not automatically mean better data. Pre-analytical variables like sample type and storage become more, not less, important as your detection limit drops, because low-level contaminants and degradation products that were invisible to a colorimetric assay suddenly show up as signal.

For most research and clinical applications, a well-optimized colorimetric sandwich ELISA with TMB detection remains the right tool. It is inexpensive, widely validated, compatible with standard plate readers, and sensitive enough for the majority of protein biomarkers. The ultrasensitive platforms earn their cost and complexity when you genuinely need to measure something at the edge of detection, like a low-abundance biomarker in early-stage disease or a trace contaminant in a large-volume sample. Matching the assay’s sensitivity to your actual biological question, rather than defaulting to the most sensitive technology available, keeps both your budget and your data interpretation grounded.