Tax incentives to encourage saving constitute an increasingly important component of government pension policy worldwide. Empirical evidence of their ability to raise household savings levels, however, is conflicting. This paper employs a semiparametric estimator to explore this problem through the study of the RRSP program in Canada. Several interesting results emerge. First, RRSP savings are found to partially, yet not fully, crowd out savings in other forms. Second, comparing different methods of controlling for observable characteristics reveals insights about the magnitude of selection bias. Third, little evidence of substitution between RRSPs and illiquid forms of savings is found.Draft, July 1998: PDF.