This paper proposes an improved version of the Akaike Information Criterion (AIC), termed AICc, for selecting the smoothing parameter in nonparametric regression. Unlike classical methods such as generalized cross-validation (GCV) and the AIC, AICc avoids the high variability and tendency to undersmooth that these methods exhibit. The authors derive AICc and compare its performance with other smoothing parameter selectors through Monte Carlo simulations. They find that AICc performs comparably well with well-behaved plug-in methods and also performs well when plug-in methods fail or are unavailable. The paper includes an application to real data and discusses potential future work, including the extension of AICc to other smoothing estimators and smoothing methods based on principles other than least squares.This paper proposes an improved version of the Akaike Information Criterion (AIC), termed AICc, for selecting the smoothing parameter in nonparametric regression. Unlike classical methods such as generalized cross-validation (GCV) and the AIC, AICc avoids the high variability and tendency to undersmooth that these methods exhibit. The authors derive AICc and compare its performance with other smoothing parameter selectors through Monte Carlo simulations. They find that AICc performs comparably well with well-behaved plug-in methods and also performs well when plug-in methods fail or are unavailable. The paper includes an application to real data and discusses potential future work, including the extension of AICc to other smoothing estimators and smoothing methods based on principles other than least squares.