ABSTRACT: Generally-speaking there are three tests for specification error in the context of regression analysis in the statistical literature: (1) The Durbin-Watson statistic, (2) The Ramsey RESET (Regression Equation Specification Error Test) t-test, and (3) The Breusch-Godfrey serial correlation LM test. The first and third tests assume that the presence of serial correlation in the regression residuals is a sign of misspecification, while the second test is more appropriate when non-linear functional forms prevail. In the RESET test, non-linearity is detected by including the squares of the fitted dependent variable in the original model, and undertaking a t-test on the coefficient of the included fitted variable. This paper introduces an alternative test for misspecification, which consists of including the squares of the actual dependent variable in the original regression. This alternative test is studied for the case of misspecification when a quadratic explanatory term is omitted. Two separate simulations are carried out. One assumes that the variables are generated by a normal distribution, and the other by a uniform distribution. The results strongly find that the alternative testing procedure suggested here dominates the other 3 tests available in the literature. This is especially true when the variables are set to follow a normal distribution.