Nonparametric tests can be made for bias in estimate and forecast errors without assuming identical and independent distributions. Tests are created for bias in the median and the mean. The test for median bias is a form of the familiar Sign Test for the median. For mean bias, an asymptotically normal test statistic is derived from the mean algebraic percentage error. When dependence has a particular structure, a variant of the jackknife may be applied to obtain an unbiased test. These statistics are then applied to cross-sectional and time series contexts.