Most times the effects of social interventions targeted to specific subgroups of the population are "a priori" uncertain. This is why measuring them "ex post" is of major value to learn which interventions work, on which target groups, in which macroeconomic conditions. The exponential growth of the literature on program evaluation over the last decade witnesses the relevance to the policy makers and to the public opinion of having available objective measures of whether a social intervention actually produced the effects it was intended to produce. In its bare essentials any evaluation strategy is based on comparing what happens to the individuals exposed to the intervention to what happens to "comparable" individuals not exposed to the intervention. The fundamental problem one needs to solve to measure the effects of the intervention is how to guarantee the comparability between the two groups of individuals. In this paper we review the recent literature on program evaluation focusing on three issues: i) the need of collecting data tailored to the specific evaluation problem at hand; ii) how parametric methods can happen to hide the lack of comparability between exposed and unexposed individuals and why nonparametric methods cannot; iii) the implications of the intervention effects being heterogeneous across individuals. Illustrations of the three issues with reference to the evaluation of Italian labour market policies are provided.