Microscopic Techniques and Criteria for Diagnosing Bacterial Vaginosis

Diagnosing bacterial vaginosis under the microscope comes down to reading the balance of bacterial shapes on a stained or unstained vaginal smear. The current gold standard is the Nugent scoring system, which grades a Gram-stained slide on a 0-to-10 scale based on three bacterial morphotypes. A bedside alternative, the Amsel criteria, relies on a wet-mount preparation and a few clinical signs. Both approaches center on the same underlying shift: the loss of rod-shaped lactobacilli and the appearance of smaller, mixed bacteria that coat vaginal epithelial cells. But the details of how each method works, where they agree, and where they fall short matter a great deal for getting the diagnosis right.

The Nugent Gram Stain

The Nugent score, introduced in 1991, was designed to make Gram stain interpretation more reproducible across different laboratories. A vaginal smear is heat-fixed, Gram-stained, and examined under oil immersion at 1,000× magnification. The reader scores three bacterial morphotypes on a 0-to-4 scale: large gram-positive rods (lactobacilli), small gram-variable or gram-negative rods (representing Gardnerella and Bacteroides-like organisms, scored together), and curved gram-variable rods (representing Mobiluncus-like organisms, weighted more heavily because of better inter-observer agreement). The three sub-scores are combined into a total between 0 and 10.¹

The total is interpreted in three bands: 0 to 3 is normal, 4 to 6 is intermediate, and 7 to 10 indicates BV.² The intermediate category is one of the system’s acknowledged weak spots. It captures women whose flora is shifting but hasn’t clearly tipped into BV, and clinicians often struggle with what to do with that result. A woman scored as intermediate today could shift to normal or to BV within days, so the score captures a snapshot that may not reflect a stable state.

What the Morphotypes Actually Look Like

Reading a Nugent smear is fundamentally a pattern-recognition task. Lactobacilli appear as long, thick, gram-positive rods, and their presence in high numbers is the hallmark of a healthy vaginal flora. When BV develops, these are replaced by a mixed population of much smaller organisms. The Gardnerella-Bacteroides group shows up as tiny gram-variable or gram-negative rods and coccobacilli, often densely packed. Early work on Gram-stained vaginal fluid established that this inverse relationship between Lactobacillus morphotypes and Gardnerella morphotypes is central to the diagnosis.³

The curved gram-variable rods deserve a closer look because they carry extra weight in the Nugent system. These correspond to Mobiluncus species, comma-shaped or “gull-wing” shaped organisms with tapered ends that appear singly or in pairs.⁴ Their presence is a strong indicator of BV, and the Nugent system was intentionally designed to weight them more heavily because observers in different labs agreed on their identification more consistently than on other morphotypes. Gram-positive cocci, by contrast, were dropped from the scoring system altogether because agreement on them was poor.¹

Amsel Criteria and the Wet Mount

The Amsel criteria take a different approach. Instead of a scored Gram stain, the clinician evaluates four signs at the bedside, three of which must be present: a thin, homogeneous vaginal discharge; vaginal pH above 4.5; a positive “whiff test” (a fishy odor when the discharge is mixed with potassium hydroxide); and the presence of clue cells on a wet-mount microscopic preparation. Of these four, the clue-cell finding is the most directly microscopic and arguably the most specific single component.

Clue cells are vaginal epithelial cells so heavily coated in adherent bacteria that the cell borders look blurred or stippled. The bacteria responsible for this coating include Gardnerella vaginalis, Mobiluncus species, Mycoplasma hominis, and Peptostreptococcus species. As a standalone finding, clue cells have been reported to have roughly 53 to 90% sensitivity and 40 to 100% specificity for BV, depending on the study and the threshold used.⁵ That wide range reflects the challenge: identifying clue cells requires training and a practiced eye, especially under standard bright-light microscopy where the bacterial coating can be subtle.

How Nugent and Amsel Compare

You might expect two validated methods for the same condition to agree most of the time, but the overlap is messier than that. When Amsel criteria are measured against the Nugent score as a reference, the reported sensitivity of Amsel varies widely. One study from Nepal found that Amsel criteria had only 50% sensitivity and 98% specificity compared to Nugent.⁶ Another found higher sensitivity at about 85%, with specificity around 96%, and noted that Amsel performed much better when the Nugent score was very high (9 or 10) than when it was borderline (7 or 8), where sensitivity dropped to about 81%.⁷

Among HIV-infected women, the agreement was especially poor: fewer than half of women who scored positive on Nugent also met the Amsel criteria.⁸ The lesson from these comparisons is that Amsel and Nugent are measuring overlapping but not identical things, and the choice of method affects who gets diagnosed. The Amsel approach depends on subjective judgment for several of its four components, which may explain why its agreement with Nugent fluctuates so much between settings.

True Clue Cells Versus Pseudo Clue Cells

One microscopic subtlety that trips up less experienced readers is the difference between genuine clue cells and what are called pseudo clue cells. True clue cells form when a biofilm grows directly on the epithelial cell surface in a cohesive, adherent pattern. When you scan the slide, you’ll see many epithelial cells with the same kind of uniformly structured bacterial coating, cell after cell. Finding at least 10 cells with this reproducible, tightly adhered pattern is considered reliable identification.⁹

Pseudo clue cells look superficially similar but arise from a different process. When bacteria are growing in a dispersed, non-adherent fashion, they can surround and secondarily embed epithelial cells. In these cases the bacterial coating on each cell looks unique and inconsistent from one cell to the next, even when many cells are covered. Recognizing this distinction matters because pseudo clue cells can lead to false-positive BV diagnoses when the underlying flora is organized differently from true biofilm-associated BV.

Phase Contrast Versus Bright Light Microscopy

The type of microscope matters more than most clinicians realize. A multicenter study comparing conventional bright-light microscopy with phase contrast microscopy found that phase contrast dramatically improved agreement between observers. For clue cell detection, inter-observer agreement under phase contrast reached kappa values of 0.69 to 0.94, while the same investigators using conventional bright light scored only 0.37 to 0.72. For lactobacillary grading, the improvement followed a similar pattern. Even intra-observer consistency improved: the same reader assessing the same slides at different times was far more reliable with phase contrast.¹⁰

Phase contrast makes translucent structures like bacteria and cell edges stand out against the background without staining, which helps readers spot the subtle coating on clue cells and distinguish lactobacilli from smaller morphotypes. In clinical settings where wet-mount preparations (rather than Gram stains) are read, switching from bright-field to phase contrast is one of the simplest upgrades available to improve diagnostic accuracy.

Observer Agreement on Nugent Scoring

Since the Nugent score depends on a human reader, how well different people agree on the same slide is a fair question. Across studies, interobserver reliability has generally been rated as good, though not perfect. One evaluation found concordant results among readers in about 64% of smears, partial agreement in roughly 32%, and true discordance in about 5%.¹¹ A separate study confirmed good overall agreement and concluded the method was reliable across different laboratory settings.¹²

Most disagreement clusters around the intermediate range (scores of 4 to 6), which is unsurprising since these smears have mixed morphotypes that don’t clearly fall into either category. At the extremes, 0 to 2 or 8 to 10, readers tend to agree. The practical takeaway is that a high Nugent score is a confident diagnosis, while a score in the middle zone warrants caution and possibly repeat testing.

The Hay-Ison System and Simpler Alternatives

Not every laboratory has the training infrastructure for full Nugent scoring. Simpler alternatives exist, the most widely referenced being the Hay-Ison system (also called Hay’s criteria or Ison-Hay grading). This method classifies a Gram-stained smear at 1,000× magnification into just three grades: Grade I (normal), with Lactobacillus morphotypes only; Grade II (intermediate), with reduced lactobacilli and mixed morphotypes; and Grade III (BV), with mixed morphotypes and few or absent lactobacilli.¹³ The grading is faster and easier to teach because it sidesteps the point-by-point morphotype quantification. The tradeoff is less granularity, which makes it harder to study transitional or borderline cases in a research context.

Self-Collected Swabs for Microscopy

A question that comes up in screening programs and research settings is whether women can collect their own vaginal swabs reliably enough for Gram stain analysis. Multiple studies suggest they can. Self-collected lower vaginal swabs have shown sensitivity for BV around 88 to 91% compared with clinician-collected specimens, with strong agreement between the two collection methods.¹⁴¹⁵ At the microbiome level, self-collected and physician-collected swabs sample the same microbial diversity.¹⁶ This matters for research cohorts and for settings where a pelvic exam is impractical or creates barriers to screening.

Conditions That Look Like BV Under the Microscope

Several vaginal conditions can mimic or overlap with BV on microscopy, and mistaking one for another leads to the wrong treatment. Two are worth knowing about.

Aerobic vaginitis (AV) shares BV’s loss of lactobacilli on a smear but has distinct microscopic features. Where BV shows a characteristic granular background of small organisms and clue cells, AV instead shows cocci or coarse bacilli, parabasal (immature) epithelial cells, and abundant white blood cells with a granular appearance.¹⁷¹⁸ White blood cells are a key distinguisher: BV, unlike most infections, typically produces little to no inflammatory cell infiltrate. AV can also co-exist with BV. Among one group of women with AV, about 37% of mixed infections involved concurrent BV.¹⁹ These mixed cases are easily missed if the reader only looks for one pattern.

Cytolytic vaginosis sits at the opposite end of the spectrum. Here, lactobacilli are actually overabundant, and their overgrowth causes lysis (breakdown) of epithelial cells. On a wet mount or Pap smear, you see excessive lactobacilli, naked nuclei from destroyed epithelial cells, minimal white blood cells, and no pathogens like Gardnerella or Candida.²⁰²¹ The vaginal pH stays acidic (3.5 to 4.5), which is the opposite of BV. Cytolytic vaginosis is more commonly confused with yeast infections than with BV, but a reader who only glances at the lactobacilli count without considering the overall pattern could miscategorize it.

Deep Learning and Automated Slide Reading

Human variability in reading Nugent smears has spurred efforts to automate the process with artificial intelligence. A convolutional neural network trained on expert-scored images achieved about 82% sensitivity and 97% specificity for identifying altered flora and BV on a validation set of nearly 6,000 images, and roughly 75% accuracy across all three Nugent categories on an independent test set, which was about 7 percentage points higher than the average of three trained technologists.²² A more recent study from Japan pushed this further, with an optimized model reaching 94% accuracy at 1,000× magnification, edging past the average 92% accuracy of the technicians it was compared against.²³

These are promising numbers, but they come with caveats. The models were trained and validated on images from specific hospital populations, and performance on slides from different settings, prepared with slightly different staining protocols, remains an open question. Still, the consistency of a machine that doesn’t get tired or lose focus after a hundred slides in a row has obvious appeal for high-throughput screening labs.

Why Standard Criteria Stumble in Postmenopausal Women

Both Amsel and Nugent were developed and validated in premenopausal women, whose vaginal flora is heavily shaped by estrogen-driven glycogen production in the epithelial cells. After menopause, especially without estrogen replacement, the baseline flora shifts substantially. Lactobacilli may be naturally scarce, the pH rises, and the epithelium thins. All of these changes can push diagnostic scores into the “BV” range in women who don’t actually have the condition.²⁴ Reported BV prevalence in postmenopausal women has ranged from 2% to 57% using traditional criteria, a suspiciously wide spread that likely reflects overdiagnosis rather than true variation. Molecular studies using gene sequencing suggest the postmenopausal vaginal microbiome is simply different in composition, and interpreting it through a premenopausal lens introduces systematic error.

Fluorescence Imaging and Biofilm Architecture

Beyond routine clinical microscopy, research tools have revealed that BV involves a structured biofilm, not just a random overgrowth. Fluorescence in situ hybridization (FISH) uses species-specific probes that glow under fluorescence microscopy, allowing researchers to see exactly which organisms are present and how they’re organized. One study using FISH showed that BV biofilms are dominated by Gardnerella vaginalis, and that Atopobium vaginae, when present, is always embedded within a Gardnerella biofilm rather than forming its own. About 41% of Gardnerella-positive samples had no Atopobium at all, but the reverse never occurred.²⁵

This kind of spatial information is invisible on a standard Gram stain, where all small gram-variable rods look alike regardless of species. FISH isn’t a routine diagnostic tool, but understanding the biofilm architecture helps explain why BV recurs so stubbornly after antibiotic treatment: killing the dispersed bacteria may leave the structured biofilm intact on the epithelial surface, ready to regrow. It also reinforces why the cohesive, uniform coating seen on true clue cells is biologically meaningful and distinct from the haphazard bacterial accumulation seen in other conditions.