Grid-by-number ("Wolman") sampling is commonly used to characterize the texture of surficial fluvial sediments. Assessments of sampling precision have concentrated on the median grain size, but adequate representation of other grain size percentiles is paramount in a number of sedimentological contexts. The field effort required to construct percentile sampling distributions by replication is prohibitive, and the absence of any general size distribution for fluvial sediments precludes theoretical calculation of percentile standard errors. Bootstrapping is used in this paper to determine percentile standard errors for large samples from two gravel bar surfaces. The improvement of percentile precision with sample size follows the theoretical expectation, hence improvement in percentile precision is achieved only at the expense of much greater sampling effort once sample sizes reach about 400 stones. However, absolute percentile standard errors differ significantly from theoretically derived values. Of particular interest is the stability of coarser percentile estimates in coarse-skewed distributions; D 95 estimates may be as precise as D 50 estimates for a given sample size. The sensitivity of the precision of estimated percentiles to distribution shape is examined using synthetic skewed and bimodal grain size distributions. The absolute precision of percentile estimates is shown to vary with the grain size distribution of a particular sediment. Our field results provide a reference for further work.