This paper proposes a nonparametric-regression-based test for omitted variables,which is applicable in both cross-sectional and time series contexts. Our test not only avoids the model misspecification problem,but also are locally more powerful than the existing tests. Moreover,unlike many other nonparametric-based tests,we use a single bandwidth rather than two different bandwidths in estimating both the conditional joint and marginal expectations,which significantly improves the size performance of our test in finite samples.Monte Carlo studies demonstrate the well behavior of our test in finite samples,which could not only capture the omitted variables feature in linear and nonlinear regressions,but also is more powerful than Ait-Sahalia et al. ’s (2001) test. In an application to testing the nonlinear Granger causality in mean,we document the existence of nonlinear relationships between theoutput gap and inflation, that is,the nonlinear“output-inflation”type of Phillips curve maybe is more suitable for China’s inflation forecast.