The article introduces new tests to check for unit roots in panels with different characteristics. They suggest a standardized t-bar test based on averaged Dickey-Fuller statistics, which converges to a standard normal distribution as the number of time series and cross-sectional units increase. When errors are uncorrelated, a modified t-bar test is distributed as standard normal for large cross-sectional units. A fixed N and T test is also developed using the average of DF statistics. Monte Carlo simulations show that with appropriate lag orders, the t-bar test performs well, outperforming a previous test by Levin and Lin.