Every ELISA fails eventually. The question is whether it fails loudly — a blank plate, a signal that fills every well — or quietly, with data that looks fine until you try to reproduce it. This guide covers the failure modes we see most often, ordered by symptom, with the most likely causes ranked by frequency and the concrete fix for each.

Bookmark this page. The next time a plate misbehaves at 6 PM on a Friday, you will want the checklist, not a textbook.

Symptom 1 — weak or no signal

The plate reads barely above blank everywhere, including the top standard.

Most likely causes, in order:

Reagent problem. Expired or degraded substrate, a detection antibody stored at the wrong temperature, or an HRP conjugate that has seen too many freeze–thaw cycles. Test the conjugate directly: add substrate to a dilution of the conjugate in a few spare wells. No color = dead conjugate.
Wrong dilution of a working reagent. Detection antibody at 1:10000 instead of 1:1000 is a classic. Recalculate every dilution from the vial label, not from your spreadsheet's memory.
Incubation too short or too cold. Binding kinetics at 4 °C are dramatically slower than at room temperature or 37 °C. A "quick" 30-minute incubation that the protocol assumes at 37 °C will underperform on the bench.
Wavelength mismatch. TMB stops at 450 nm. PNPP reads at 405 nm. Check that the reader filter matches the substrate you actually used.
The analyte is genuinely absent or degraded. If only samples are weak but the standards are fine, the assay works — your samples don't. Check storage history and freeze–thaw count.

Symptom 2 — high background everywhere

Blank wells read 0.2–0.4 OD and the whole plate looks "muddy".

Most likely causes:

Insufficient washing. The single most common cause of high background. Wash more times (4–6), with longer soak times (30–60 s per cycle), and verify the washer actually dispenses and aspirates every well — a clogged pin produces stripes of high background.
Cross-reactivity of the detection antibody. Run a "no-analyte, full reagents" control. If the blank is high with all reagents present but clean without the detection antibody, the detector is binding non-specifically.
Contaminated substrate or buffers. Substrate that has been exposed to light or to trace metal contamination develops color on its own. Use fresh aliquots.
Insufficient blocking. Extend blocking time, increase blocker concentration, or switch blocker (BSA vs. casein vs. commercial formulations behave differently per antibody pair).
Matrix effect in samples only. If standards are clean and samples are high, the problem is the matrix — see our guide on spike recovery and dilution linearity.

Symptom 3 — high CV between replicates

Duplicates disagree by 15 %, 30 %, sometimes more. The target is CV ≤ 10 % for most of the curve range (≤ 15–20 % at the LLOQ in regulated work).

Most likely causes:

Pipetting inconsistency. The biggest contributor by far. Use calibrated pipettes, reverse pipetting for viscous samples, and never pipette volumes below 2–5 µL directly into wells.
Incomplete washing or residual liquid. Droplets left after the final wash dilute the substrate unevenly. Tap the plate firmly on absorbent paper after the last aspiration.
Edge effects — see the next section.
Bubbles in wells. Even one bubble distorts the optical path. Inspect the plate before reading and pop bubbles with a clean needle or a quick centrifuge pulse.
Analyte instability in the dilution series. If high CV appears only at specific concentrations, the analyte may be adsorbing to tubes or precipitating. Use low-bind tubes and prepare dilutions fresh.
Reading too late. Signal drifts after the stop reaction. Read within the window the kit specifies.

A structured way to handle genuine outliers after all of this is ruled out: run replicates at least in duplicate, flag pairs above your CV threshold, and only then consider formal outlier tests (Grubbs, Dixon) — never delete a replicate just because it is inconvenient. Labreador's group comparison module automates CV flagging and significance testing so the decision is documented, not arbitrary.

Symptom 4 — edge effects

Outer rows and columns read systematically higher (or lower) than the plate interior. The classic pattern: row A and row H both elevated.

Causes and fixes:

Temperature gradients. The plate edge equilibrates with room air faster than the center. Incubate the plate in a humidified chamber or a sealed box, and never on a cold metal surface.
Evaporation. Long incubations without a plate sealer dry the outer wells first. Always seal, and consider skipping the outermost wells for critical assays (use them for buffer or leave them empty).
Uneven washing. Some washers aspirate edge wells differently. Run a dye test to verify.

Quick diagnostic: plot mean OD by row and by column. A U-shaped (or inverted-U) profile confirms an edge effect rather than random noise.

Symptom 5 — the standard curve looks wrong

The curve is flat, saturated, hook-shaped, or the software refuses to fit.

Flat curve (low ΔOD): the dynamic range between your top and bottom standard is too small. If the difference between the highest and lowest standard is below ~0.5 OD, the fit becomes unstable and EC50/IC50 confidence intervals explode. This usually traces back to symptom 1 (weak reagents) or an incorrect standard stock.
Saturated top of the curve: signal exceeds the reader's linear range. Shorten substrate incubation or read earlier.
Hook shape at high concentration: the prozone/hook effect in one-step sandwich assays. Dilute the top samples — the true concentration is higher than the reading suggests.
Standards out of order or a swapped dilution: back-calculate each standard against the fitted curve. A standard whose recovery is off by 50 % while its neighbors are fine is almost certainly a pipetting error in the dilution series, not a fitting problem. Exclude it, refit, and document the exclusion.

This is exactly what the Method Health checks in Labreador's ELISA module are for: they automatically report standard back-calculation recovery, ΔOD between top and bottom standards, EC50 inside or outside the calibrated range, and unstable confidence intervals — so a bad curve is flagged before you report concentrations from it. And if you are still fitting log-linear regressions instead of a proper 4PL/5PL, read why 4PL fits ELISA better and why R² alone is not enough.

Symptom 6 — results don't reproduce between plates

Today's curve and yesterday's curve give different concentrations for the same control sample.

Track a control sample on every plate. Without an inter-plate control you cannot distinguish assay drift from sample differences.
Check the curve parameters, not just R². If the upper asymptote or the slope factor moves substantially plate-to-plate, something in the reagents or incubation is drifting — temperature, timing, or reagent age.
Standardize read timing and temperature. A 10-minute difference in substrate incubation shifts the whole curve.
Watch lot changes. New antibody or standard lot = re-validate, at least with a bridging run of old vs. new.

A 60-second triage checklist

When a plate looks wrong, answer these in order before repeating anything:

Do the blanks look normal? (If not → background/washing problem.)
Do the standards span a normal signal range and shape? (If not → reagent, dilution series, or reader problem.)
Do replicates agree? (If not → pipetting, washing, or edge effect.)
Do back-calculated standards recover within 80–120 %? (If not → curve or dilution problem.)
Does the control sample match its historical value? (If not → drift between runs.)

Five questions isolate roughly 90 % of ELISA failures into the right category, and each category maps to one of the sections above.

Analyzing problem plates in Labreador

Labreador runs the entire diagnostic part in your browser, with no data upload:

The ELISA module flags high-CV replicate pairs, reports per-standard recovery, and warns when results sit outside the calibrated range instead of silently extrapolating.
Method Health checks catch low dynamic range, drifting curve parameters and unstable confidence intervals automatically.
The quality gate scores each fit beyond a naked R², so marginal curves are visible before they reach your report.

Raw plate data never leaves your device.

References

Aydin, S. (2015). A short history, principles, and types of ELISA, and our laboratory experience with peptide/protein analyses using ELISA. Peptides, 72, 4–15.
Tighe, P. J., et al. (2015). ELISA in the multiplex era: potentials and pitfalls. Proteomics Clinical Applications, 9(3–4), 406–422.
Findlay, J. W. A., & Dillard, R. F. (2007). Appropriate calibration curve fitting in ligand binding assays. The AAPS Journal, 9(2), E260–E267.

Cite Labreador

If Labreador supported your analysis, please cite it: Labreador - Bioassay Analysis Platform (Version 1.6.0) [Computer software]. Zenodo. https://doi.org/10.5281/zenodo.21676846

ELISA Troubleshooting Guide: Weak Signal, High Background, High CV and Other Common Problems