This study evaluates the impact of bike-sharing programs on urban traffic congestion. Utilizing the universe of Lyft Divvy bike-sharing trips in Chicago up to 2019, I estimate the causal effect of increasing the number of Divvy trips on rush-hour congestion on the spatial level of census tracts. While ordinary least squares regressions with two-way fixed effects reveal no significant effect, the nonparametric treatment-response curve based on generalized propensity score matching suggests that average traffic speed in a census tract responds little to less than 70 Divvy trips in the area but exhibits a substantial 20% increase as Divvy trips increase to 100. This suggests a threshold effect in the efficacy of bike-sharing to alleviate congestion.