Recent models have shown that the development of spatial structure in plant mixtures may make strong competitive interactions between species hard to detect owing to spatial segregation of the competing species. Here we address the issue of measuring interspecific competition using a simulation based on a neighbourhood population model which assumes that both dispersal and competitive interactions are localized. Using known parameter combinations we use the model to test the power and efficiency of two approaches for detecting and measuring competition. The first approach is based on measuring the response of communities to the removal of neighbours. Measures of interspecific competition based on this approach are extremely biased by spatial segregation of species, although this bias may be partially overcome by altering the spatial scale at which the effects of removals are recorded. The second approach is based on multiple regression of per capita population growth rates on local densities of the interacting species. When dispersal is restricted, the regression approach provides accurate estimates of interspecific competition coefficients when the scale of the sampling unit (i.e. the quadrats within which plants are counted) is large compared to the scale at which interactions and dispersal occur. When seeds disperse globally the removal method performs best; the regression method fails because sampling units do not form closed dynamic systems. Our results highlight the importance of tailoring methods for detecting competition to the characteristics of the species in question. They also indicate that rapid nonmanipulative estimates of competition coefficients may be the best approach in communities where dispersal is restricted and competitive interactions localized, which is likely to be the case for the majority of plants.