When considering splitting behaviour into bouts, it is best to display the data in log frequency rather than in log survivorship plots. While the latter has the advantage of simplicity, the former has the advantage that the data points are independent of each other, so that a non-linear curve-fitting procedure, such as NLIN in SAS, can be applied. Traditional analysis of variance can then be used to decide whether the behavioural events are split into bouts, and least-squares estimates and standard errors can be calculated for each parameter of the fitted model. These estimates can be used in Slater & Lester's (1982 Behaviour, 79, 153–161) formula to calculate the best criterion interval to use in splitting behaviour into bouts. Average bout length and number of bouts are, however, best estimated directly from the model's parameters.