In this paper we test the convergence hypothesis by using a revised 4- step procedure of panel unit root test suggested by Evans and Karras (1996). We use data on output for 24 OECD countries over 40 years long. Whether the convergence, if any, is conditional or absolute is also examined. According to a proposition by Baltagi, Bresson, and Pirotte (2005), we incorporate spatial autoregressive error into a fixedeffect panel model to account for not only the heterogeneous panel structure, but also spatial dependence, which might induce lower statistical power of conventional panel unit root test. Our empirical results indicate that output is converging among OECD countries. However, convergence is characterized as conditional. The results also report a relatively lower convergent speed compared to conventional panel studies.