Human expertise is a significant source of information about environments with inherent uncertainty. However, it is well documented that subjective estimates of chance tend to violate the mathematical axioms of probability, that is, they are incoherent. This fact makes the use of such estimates problematic for statistical inference, decision analysis, economic modelling or aggregation of expert opinions. In order for the subjective probability estimates to be used in a correct and meaningful way, they must be reconstructed so that they are coherent. The proposed algorithms for coherent reconstruction are based on heuristic search methods, namely, Genetic Algorithms and Simulated Annealing. These algorithms are combined with efficient data structures that compactly represent probability distributions. The reconstructed estimates are coherent and close to the initial judgments with respect to some distance measure, maintaining the insight of the expert. Empirical studies shown that the coherent approximations are more stochastically accurate than the original subjective estimates.