Normal approximation to the distribution of the estimated process capability index _??__pk is discussed. The direct normal approximation to _??__pk and an alternative approach are compared in terms of skewness and kurtosis. It is shown that the latter approach improves on the former one.