The non take-up of social assistance benefits due to claim costs may seriously limit the anti-poverty effect of these programs. Yet, available evidence is fragmented and mostly relies on interview-based data, potentially biased by misreporting and measurement errors on both benefit entitlement and income levels used to assess eligibility. In this paper, we use Finnish administrative data to compare eligibility and actual receipt of social assistance by working-age families during the post-recession period (1996-2003). Possible errors due to time-period issues and discretionary measures by local agencies are carefully investigated. Non take-up is found to be substantial – between 40% and 50% – and increasing during the period. Using repeated cross-section estimations, we identify a set of stable determinants of claiming behavior and suggest that the increasing trend is mainly due to a composition effect, i.e. a decline in the proportion of groups with higher claiming propensity. We finally discuss the targeting efficiency of the social assistance scheme.