The occurrence of unexpected mean values in statistical analyses of experimental data, known as Peelle's pertinent puzzle in nuclear data evaluation, is revisited. It is shown in terms of Bayesian statistics, it is not caused exclusively by nonlinearities but is due to improper estimates of covariance matrices of experiments. Applying the correct covariance matrix leads to the exact posterior expectation value and variance for an arbitrary number of uncorrelated measurement points that are normalized with the same quantity.