The paper considers the variance of experimental estimates of the amplitude probability density function due to finite record length. Analysis, assuming a normally distributed signal, is presented for both a sampled record appropriate for digital experimental procedures and for the case of a continuous record of relevance to analogue analysis. Particular attention is given to the influence of the autocorrelation function of the examined signal on the variance of the estimate. The results may be useful in the design of experiments to estimate the amplitude probability density function, particularly with regard to the selection of the intersample interval in the case of sampled records. Some experimental results are compared with theoretical predictions.