The 1996 PRWORA reform introduced time limits on the receipt of welfare in the United States. We use variation by state and across demographic groups to provide reduced form evidence showing that such limits led to a fall in welfare claims (partly due to "banking" benefits for future use), a rise in employment, and a decline in divorce rates. We then specify and estimate a life-cycle model of marriage, labor supply and divorce under limited commitment to better understand the mechanisms behind these behavioral responses, carry out counterfactual analysis with longer run impacts and evaluate the welfare effects of the program. Based on the model, which reproduces the reduced form estimates, we show that among low educated women, instead of relying on TANF, single mothers work more, more mothers remain married, some move to relying only on food stamps and, in ex-ante welfare terms, women are worse off.