In the paper, we use a panel data approach to study the threshold effects and nonlinearity in economic growth instead of the cross-sectional approach commonly used in previous studies. The methodologies developed by Hansen (2000) and Caner and Hansen (2004) are applied to identify threshold variables. The analysis reveals many threshold variables, in particular for short lengths of panels. However, the results vary according to the number of observations included. To reduce the large number of potential threshold variables, principle component analyses are used to create a smaller number of composite factors and test the threshold effects. Using a five-year panel, we show how to classify countries into different growth typologies. Finally, a local linear nonparametric estimation method is presented to illustrate the nonlinear curves of convergence relative to the threshold variables.