96-well plate layout: design principles that protect your data

How to arrange samples, controls, and replicates to minimise systematic bias before you pipette a single well

Why layout matters

A poorly arranged plate can introduce systematic error that no statistical test will ever recover. Edge effects, temperature gradients, and pipetting drift are all spatially correlated, meaning where a well sits on the plate predicts its noise as much as its biology does.

Good layout design isn't about aesthetics. It is about making sure your statistical model can tell the difference between signal and plate position.

Controls first

Before placing any samples, fix your control positions. Every plate needs at minimum:

Control typePurposeRecommended position
Negative control (vehicle only)Sets the baseline signal floorColumn 2
Positive control (saturating compound)Sets the ceiling; anchors normalisationColumn 11
Blank (no cells or reagent)Detects background fluorescenceA few interior wells

Grouping each control in a dedicated side column (rather than interleaving it among samples) is the practical convention for hand and automated pipetting. But it comes with a real trade-off: those side columns are close to the high-variability outer ring (see Edge effects below), and your controls are exactly the wells you can least afford to be noisy because they anchor normalisation. Two ways to manage that tension:

  • Keep the outermost ring (columns 1 and 12, rows A and H) as a sample-free buffer moat and put your controls one column in, at columns 2 and 11. This is the layout the table above assumes.
  • If you must use every column and controls end up on the edge, replicate them generously and apply a spatial correction (a B-score / median-polish normalisation) so edge bias in the controls doesn't propagate into every well → normalising readings.

Whatever you choose, don't bury controls in the four corners: those are the single most evaporation- and handling-prone wells on the plate.

Always run controls in duplicate at minimum. A single outlier control well will corrupt your entire plate's normalisation.

Replicate strategy

Technical replicates

Run each sample in at least triplicate. Place replicates in non-adjacent wells (ideally in different rows) so that a pipetting error or bubble affects only one of the three.

Biological replicates

Biological replicates belong on separate plates, not in extra wells on the same plate. Same-plate replicates share all the systematic error sources and are not independent.

Randomisation

Completely random layouts are rarely practical, but partial randomisation is always achievable. At minimum:

  • Do not place all samples from one condition in the same row or column
  • Alternate sample groups across rows if a full random design is not feasible
  • Record the layout digitally before you start, since retrofitting a layout to an already-run plate is error-prone

Avoid placing high-value or irreplaceable samples in edge wells A1, A12, H1, and H12. These four corners are most exposed to evaporation and handling vibration.

Edge effects

The outer ring of wells (row A, row H, column 1, column 12) consistently shows higher signal variability due to evaporation and temperature exposure. Two practical mitigations:

  1. Fill all outer wells with buffer, even if they carry no sample
  2. Apply a plate-position correction model if edge bias is unavoidable

Toggle between a risky and a protected layout to see why control placement matters: controls stranded on the edge bias every normalised value:

Caveats

Sometimes, edge effects might not be that prominent in your system. This is dependent on how well you know your cells, and whether your readouts are affected by potential edge effects. It is always good to run pilot experiments to validate experimental conditions prior to running your main experiment.

ABCDEFGH123456789101112
Positive controlNegative controlSampleBlank / buffer

Controls fill the outer columns, where evaporation and thermal gradients hit hardest: 16 control wells on the edge. A biased control biases every normalised value on the plate.

Documenting your layout

A layout is only useful if it is recorded correctly. At minimum capture: well ID, sample ID, condition, concentration, and replicate number. Platelet reads this map directly from your uploaded file and links it to your readings automatically.

Ready to run this analysis on your own data?