The problem of minimizing the cost of evaluating a discrete function by sequentially reading its variables is a problem that arises in several applications, among them automatic diagnosis design and active learning. In this problem, each variable of the function is associated with a cost, that we have to pay in order to check its value. In addition, there may exist a probability distribution associated with the points where the function is defined. Most of the work in the area has focussed either on the minimization of the maximum cost or on the minimization of the expected cost spent to evaluate the function. In this dissertation, we show how to obtain an Ômicron logarithm of N approximation with respect to the worst case minimization (the best possible approximation under the assumption that P is different from NP). We also show a polynomial time procedure for evaluate a function that simultaneously optimizes both the worst and the expected costs.