A log backtransformed variogram was used to provide a robust estimation of the experimental variogram in the presence of highly skewed data, such as those obtained in fisheries acoustic surveys. Simulations based on North Sea herring surveys were carried out which found no evidence of greater bias in the log backtransformed variogram relative to the classical variogram. The log backtransform was then applied to six North Sea herring acoustic surveys to determine how best to utilise the available datasets. The exclusion of intertransect data and use of along-the-track data did not confer any significant additional variographic stability. When divided into eastern and western regions, values of spatial dependence (nugget as a proportion of sill) were consistently greater in the western region. The recommended dataset selection criteria is, therefore, to use the log backtransform on the whole data set (i.e. including intertransects) separated by region (east and west). Variogram modelling and the resulting inferences on herring distributions are discussed.