Information on velocity fields in rivers is critical for designing infrastructure, modeling contaminant transport, and assessing habitat. Although non-contact approaches to measuring flow velocity are well established, these methods assume a stationary imaging platform. This study eliminates this constraint by introducing a framework for moving aircraft river velocimetry (MARV). The workflow takes as input images acquired from an airplane and involves orthorectification, frame overlap analysis, image enhancement, particle image velocimetry (PIV), and aggregation of the resulting velocity vectors onto a prediction grid. We also use new metrics to quantify the agreement between image-derived and field-measured velocity vectors in terms of both orientation and magnitude. The potential of MARV was evaluated using data from two Alaskan rivers: a large, highly turbid channel and its smaller, clearer tributary. Sediment boil vortices on the mainstem provided natural features trackable via PIV and estimated velocities corresponded closely with field measurements (R² up to 0.911). We compared an exhaustive approach that evaluates overlap for all frame combinations to a simpler rolling window implementation and found that the more efficient algorithm did not compromise accuracy. Sensitivity analysis suggested that the method was robust to window parameterization. Comparing PIV output from different flying heights and imaging systems indicated that larger pixels led to higher accuracy and that a more advanced dual-camera system provided superior performance. Results from the tributary were less encouraging, presumably due to a lack of trackable features in visible images. Testing across a range of rivers is needed to assess the generality of MARV.