This paper studies alternative approaches to consider time and spatial dependence in aggregate panel models where neither N nor T is very large, a problem that applied researchers often face when working with country- or state-level panel data. I show that the variance of the two-way cluster standard errors (2CCE) is affected by both types of dependence. Therefore, when neither N nor T is large, these standard errors could be poorly estimated and thus unreliable. As a consequence tests based on the 2CCE might perform poorly in terms of rejection rates, even in panels with a moderate sample size. I show that the cluster estimator can be expressed as a fully flexible panel version of the spatial autoregressive model. Then, it is possible to exploit that insight and use some prior knowledge about the dependence to estimate more parsimonious models that deliver more precise standard errors. In a calibrated Monte Carlo exercise using state minimum wage data, I show that using a panel version of the spatial model yields substantially better results than using the 2CCE when N and T are as small as 50 and 30. Finally, I study the implications of considering both types of dependence within a state-year panel data of wage inequality and minimum wages in the US. When both types of error dependence are considered, the marginal effect of the minimum wage over wage inequality is no longer significant.