Interspecific variation in water-induced fluctuations in stem girth demonstrates the mechanisms promoting coexistence in seasonally dry tropical forest. In addition, these fluctuations are a potential, but unevaluated, source of bias in measurements of annual tree growth rates. To examine diurnal and seasonal patterns of stem diameter change, tree girth was measured over 2 years (1997–1999), using dendrometer bands, for three species ( Celtis mildbraedii, C. zenkeri and Strombosia glaucescens) in semi-deciduous forest in Ghana. Soil matric potential was measured concurrently at 15 cm depth. In addition, measurements of all trees >20 cm dbh on three, 1 ha plots were made at the beginning and middle of the 1998/1999 dry season. During the severe 1997/1998 dry season, soil matric potential declined below −1.5 MPa and two species showed significant stem shrinkage. For the evergreen species, C. mildbraedii, there was a significant positive effect of tree diameter on stem shrinkage, and shrinkage was greater in the second, compared to the first, half of the dry season. For the deciduous species, C. zenkeri, shrinkage was reduced during the second half of the dry season, following leaf fall. During 1998/1999, soil matric potential, did not decline below −1.5 MPa, and rates of girth change remained positive for all species. There were no significant effects of size or phenology on the rate of girth change in the plot-based study. Deviations in annual increment calculated over successive monthly intervals indicate that a 10-fold difference in soil water availability between measurement occasions can lead to a 4% bias in estimates of annual growth. Measurements of forest plots should be made when inter-annual variation in soil water availability is low. In this forest, measurements should, therefore, be made during the wet season, contrary to published recommendations.