Using the marginal likelihood of the residual, we propose new score type unit root tests for heterogeneous panels. Our tests are more powerful than the t-bar test of Im et al. (2003) for panels with many cross-sectional units and short time spans.