Recovery tests are based on estimating transmissivity, T, from the heads that rebound after pumping has stopped. Recovery tests can be performed in wells where conventional constant-rate pumping tests would not be possible. Test interpretation is based on the simple Theis recovery method, related to late time drawdown in an infinite homogeneous aquifer. Yet, field data often cannot be explained by the homogeneous theory. Because T is heterogeneous over an evolving range of scales, it is important to evaluate the support scale of hydraulic tests. Numerical simulations are performed to show that heterogeneity in T can explain these field observations. It is also shown that the local T value around the well can be inferred from early time-recovery data, assuming ideal conditions, whereas late time data yield a large-scale (regional) representative value. Even when recovery is observed for a short time, indirect information about the regional value can also be obtained. A method for the interpretation of recovery tests is proposed based on the Theis recovery method that takes into account the heterogeneity of aquifers. Finally, some guidelines are provided for best test performance depending on the scale of the problem.