We derive expressions for the first-order bias of the MLE for a Poisson regression model and show how these can be used to adjust the estimator and reduce bias without increasing MSE. The analytic results are supported by Monte Carlo simulations and three illustrative empirical applications.