The article presents a method for computing maximally general minimal length probabilistic decision rules from probabilistic decision tables using the idea of decision matrices. The method reduces the problem of rule computation to the problem of identification of prime implicants of a suitably defined Boolean function. The presentation is developed in the context of the variable precision rough set model, which underlies the methodology of probabilistic decision table acquisition from data. A computational example is included to illustrate the application of the method to a real-life problem.