Multivariate analytical models are quite successful in explaining one or more response variables, based on one or more independent variables. However, they do not reflect the connections of conditional dependence between the variables that explain the model. Otherwise, due to their qualitative and quantitative nature, Bayesian networks allow us to easily visualize the probabilistic relationships between variables of interest, as well as make inferences as a prediction of specific evidence (partial or impartial), diagnosis and decision-making. The current work develops stochastic modeling of the leaching phase in piles by generating a Bayesian network that describes the ore recovery with independent variables, after analyzing the uncertainty of the response to the sensitization of the input variables. These models allow us to recognize the relations of dependence and causality between the sampled variables and can estimate the output against the lack of evidence. The network setting shows that the variables that have the most significant impact on recovery are the time, the heap height and the superficial velocity of the leaching flow, while the validation is given by the low measurements of the error statistics and the normality test of residuals. Finally, probabilistic networks are unique tools to determine and internalize the risk or uncertainty present in the input variables, due to their ability to generate estimates of recovery based upon partial knowledge of the operational variables.