Simple conditions for the inconsistency of Pearson’s chi2 test in case of very sparse categorical data are given. The conditions illustrate the phenomenon of “reversed consistency”: the greater deviation from the null hypothesis the less power of the test.