Over the past years, the field of small unmanned aerial vehicles has grown significantly and several applications have appeared, requiring always more autonomous flight. An important remaining challenge for fully autonomous unmanned aerial vehicles is collision avoidance between aircraft. In this work, we will compare two collision avoidance algorithms in terms of performance and robustness to sensor noise. We will leverage both experiments with real vehicles and calibrated, realistic simulations to get an insight of the effect of noise on collision avoidance. Our results show that although algorithms that use velocity as input are better in minimizing velocity variation and generally produces more efficient trajectories, they are less robust to perception noise. On the other hand, position-based algorithms that typically generate slower and longer avoidance maneuvers, become competitive at high levels of sensor noise.