Stochastic computation of statistical moments and related quantities, such as the mean, variance, skewness and kurtosis, is performed with simple neural networks. The computed quantities can be used to estimate the parameters of input data probability distributions, gauge the normality of data, add useful features to the inputs, preprocess data and for other applications. Such neural networks can be embedded in larger ones that perform signal processing or pattern recognition tasks. Convergence to the correct values is demonstrated with experiments.