Correspondence analysis (CA) is frequently used in the interpretation of palaeontological data, but little is known about the minimum requirements for a result to be valid. Far from being a fundamental mathematical study of CA, this paper aims to present a tool, which may serve to evaluate results obtained in (palaeontological) praxis. We created matrices of random data, grouped by matrix size and varying percentages of zero cells. Each matrix was submitted to CA. Per matrix group the minimum, mean and maximum percentages of total inertia were calculated for the first four axes. We compared these results with several real cases in vertebrate paleontology. Valid conclusions based on CA can only be drawn on percentages that are considerably higher than the axis percentages obtained from random matrices.