Growth of individuals can be quantified by repeated measurements of size over time. At the population level, growth varies among-individuals, and to estimate this variation accurately, we must separate variation among individuals from problems in the measurement process that lead to errors and uncertainty. We propose a model of  measurement error of tree stem girth based on duplicate measurements of the same stems by two teams within one or two days. We found that the absolute error increased, whereas relative error decreased with stem size. To reduce such heteroscedasticity, we suggest to square-root transform stem sizes. Forest plot data show decreasing variation in growth rate with longer inter-census interval, which partly reflects the diminishing effect of measurement errors. We examine this effect in artificial populations, and propose a procedure for estimating individual growth parameters from census data when census intervals vary.