Here, we promote the perspective that a computational model can be a rigorous crystallization of a hypothesis for the mechanisms generating observed data. We provide an example of using this approach to discriminate among hypotheses despite uncertainty in parameter values. Humans have been shown to produce non-uniform patterns of force fluctuation when they exert force in different directions with the index finger. We computationally formulated two hypotheses for this observation based on different cost functions of muscle effort, and then stochastically explored the space of unknown parameters to convergence to generate probability distributions of predictions from each hypothesis. The observed data were not within the probability distribution for Hypothesis 1: the sum of muscle forces is minimized, but were within the corresponding distribution for Hypothesis 2: the sum of squared muscle forces is minimized. Therefore, this approach provides rigorous evidence that Hypothesis 2 can not be rejected in favor of Hypothesis 1. The advantages and pitfalls of this computational approach to hypothesis testing are discussed.