Predicting the post-impact velocity for torquecontrolled flexible-joint robots allows the implementation of impact-aware control schemes which exploit intentional collisions for robotic manipulation and locomotion. Starting from the existing rigid-robot post-impact velocity prediction approach [1], this paper shows how an improvement in the prediction quality can be obtained by taking into account impact surface friction, joint motor inertias, gear ratios, and low-level joint torque control gains. Compared to the previous rigid-robot approach, the paper also proposes a more robust method to estimate the gross post-impact velocity profile from experimental data via a polynomial fit, to remove unmodelled and secondary vibratory effects. The approach is illustrated by means of simulation and validated on 50 experimental trials on a commercially available torque-controlled robot. The recorded impact data and prediction algorithms are shared openly for reproducibility.