We describe a method of producing small-area estimates of characteristics of the neighbourhood social environment, such as participation in associational life. The method may be characterised as ‘modelling nationally, and predicting locally’ because the initial models are estimated using multilevel data derived from national datasets, calibrated to allow for within-region as well as between-region variation. These models produce coefficients which can then be applied to socioeconomic data at the local scale. As an example we outline a model of the determinants of volunteering. This indicates significant variation between places in the relationship between individual and area characteristics which influence the probability of volunteering. Then we proceed to show how this can be used to generate predictions of the variations between places in a number of aspects of the social environment. The estimates we produce are validated against survey data from other surveys, and a satisfactory degree of correlation is obtained, giving us confidence in the estimates we have produced.