A new adaptive mesh refinement strategy is presented that couples feature-detection with local error-estimation. The goal is to guide refinement to key vortical features using feature detection, and to terminate refinement when a maximum acceptable error level has been reached. The feature detection scheme, which has been presented in previous related work, uses a special local normalization that allows it to properly identify regions of high vortical strength without tuning to a particular vorticity value. The newly introduced error estimation scheme applies a Richardson extrapolation-like procedure to detect local truncation error based on solutions from different grid levels. The error is then used the computed error to determine when to cut off further refinement. The paper presents a theoretical analysis of the scheme, applying it to computations of an isolated vortex and comparing to an exact solution. The scheme is implemented as part of the off-body Cartesian solver in the Helios code. Two practical cases are considered, resolution of the wake tip vortex from a NACA 0015 wing, and resolution of the wake structure of a quarter-scale V22 rotorcraft.