Rock abrasivity is the main factor that causes erosion of excavation tools and is usually quantified by the Cerchar Abrasivity Index (CAI). Although Cerchar abrasivity tests are easy to perform, they are time consuming and require a relatively high volume of rock samples. Having good correlations of CAI values and other faster and simpler tests is therefore of great interest, since it results in time and budget savings when controlling excavating tool wear. Based on the results of 73 andesitic rock samples coming from the central area of Ecuador, this paper presents a series of artificial neural networks developed to find a good estimation of CAI values of andesitic rocks from their petrographical properties. The network showing the best performance (R2 equal to 97%) is identified and a detailed process to estimate CAI value using the network developed is described.