We performed quantitative comparisons with the two-dimensional gel electrophoresis technique and evaluated the reliability of biostatistical tests for the correction of "false significant" results (alpha-error) by performing repeated runs of an experiment. Results based on uncorrected p-values yielded numerous significant differences in spot intensity which could not be replicated in two additional runs. The best strategy for avoiding these "false-positive" results was strongly dependent on the type of result. In experiments yielding very marked group differences in spot intensity, calculation of the "False Discovery Rate" (FDR) by the Benjamini and Hochberg method corrected the results with sufficient reliability. In experiments yielding relatively small (p-values>0.001) group differences, up to 100% of all results which were significant in two repeated runs were excluded ("false-negative") by calculation of the FDR. In such experiments, significant differences need confirmation by repeated runs.