Post stratification is usually judged in the context of the variance of the post stratification estimator taken over all possible sample configurations appropriately weighted by the probability of occurrence. The generally accepted view is that, in most cases, the technique has little to offer over using the sample mean. This basis for evaluating the technique is questioned and the appropriate framework for statistical inference discussed. It is argued that inferences should be made conditional on the achieved sample configuration. The theory developed shows that neither the post stratification estimator nor the sample mean is uniformly best in all situations but empirical investigations indicate that post stratification offers protection against unfavourable sample configurations and should be viewed as a robust technique.