The technical and scale efficiency of Polish farms is analysed, using Data Envelopment Analysis. Efficiency differences are measured according to farm specialisation, in crop or livestock, at two points in time during transition, 1996 and 2000. The statistical variability of efficiency estimates is investigated. The efficiency results are reviewed in the light of confidence intervals provided by bootstrapping and of a summary measure introduced in this study 'the coefficient of separation'. The inference analysis suggests that farms might be less efficient than revealed by the point estimates alone, and that they might not be clearly different from each other.