This paper investigates the effect of semi-liberty as an alternative to prison on recidivism in France. Our analysis is based on a unique dataset comprising 1,445 offenders, all eligible to semi-liberty. In the absence of an instrumental variable affecting access to semi-liberty but unrelated to recidivism, we turn to selection-on-observable methods as well as sensitivity analyses to bound the causal effect of interest. Our results under treatment exogeneity (Cox regressions) and conditional independence (entropy balancing matching) show that semi-liberty is associated with a reduction of 22% to 31% in offenders’ hazard of recidivism in the five years after release. The estimated effects decrease, but remain negative and significant when credible confounders are introduced. Overall, our analysis lends strong support for a beneficial effect of semi-liberty compared to prison.