This paper tests for discrimination against immigrant defendants in the criminal justice system in Chile using a decade of nationwide administrative records on pretrial detentions. Observational benchmark regressions show that immigrant defendants are 8.6 percentage points less likely to be released pretrial relative to Chilean defendants with similar proxies for pretrial misconduct potential. Diagnostics for omitted variable bias -- including a novel test to assess the quality of the proxy vector based on comparisons of pretrial misconduct rates among released defendants -- suggest that the discrimination estimates are not driven by omitted variable bias and that, if anything, failing to fully account for differences in misconduct potential leads to an underestimation of discrimination. Our estimates suggest that discrimination stems from an informational problem because judges do not observe criminal records in origin countries, with stereotypes and taste-based discrimination playing a role in the problem's resolution. We find that discrimination is especially large for drug offenses and that discrimination increased after a recent immigration wave.