This study investigates multisensory integration in weakly electric fishes, Apteronotus albifrons and Eigenmannia virescens, during their natural refuge tracking behavior. Our goal is to examine and model the sensory reweighting that occurs in the midbrain in relation to changes in the sensory salience. Specifically, we manipulate illumination (light, dim light, dark), conductivity (low, medium, high), refuge structure (with/out windows), and refuge length (short, medium, high) and observe how the weights of visual and electrosensory systems are dynamically adjusted. We conducted a total of 2700 trials with N=10 individuals under different sensory salience conditions. We used the minimum-variance unbiased estimator (MVUE) model to capture the characteristics of the sensory reweighting process. The stochastic nature originating from the trial-to-trial variability in behavioral responses makes it difficult to assess the predictive power of the MVUE model. To remedy this problem, we calculated the natural variability in the behavioral response distributions of the fish. We used this variability range to decide if the model predictions can be distinguished from the actual fish trajectories or not. In addition, we compared the model predictions against the distributions yielded by random weights. Our statistical analysis showed that the predictions of the MVUE model are significantly different than that of the random. We also conducted systematic cross-validation experiments to demonstrate the generality of our results under different sensory salience conditions.