We typically slow down after committing an error, an effect termed post-error slowing (PES). Traditionally, PES has been calculated by subtracting post-correct from post-error RTs. Dutilh et al. (Journal of Mathematical Psychology, 56(3), 208-216, 2012), however, showed PES values calculated in this way are potentially biased. Therefore, they proposed to compute robust PES scores by subtracting pre-error RTs from post-error RTs. Based on data from a large-scale study using the flanker task, we show that both traditional and robust PES estimates can be biased. The source of the bias are differential imbalances in the percentage of congruent vs. incongruent post-correct, pre-error, and post-error trials. Specifically, we found that post-correct, pre-error, and post-error trials were more likely to be congruent than incongruent, with the size of the imbalance depending on the trial type as well as the length of the response-stimulus interval (RSI). In our study, for trials preceded by a 700-ms RSI, the percentages of congruent trials were 62% for post-correct trials, 66% for pre-error trials, and 56% for post-error trials. Relative to unbiased estimates, these imbalances inflated traditional PES estimates by 37% (9 ms) and robust PES estimates by 42% (16 ms) when individual-participant means were calculated. When individual-participant medians were calculated, the biases were even more pronounced (40% and 50% inflation, respectively). To obtain unbiased PES scores for interference tasks, we propose to compute unweighted individual-participant means by initially calculating mean RTs for congruent and incongruent trials separately, before averaging congruent and incongruent mean RTs to calculate means for post-correct, pre-error and post-error trials.