The determination of soil shear strength parameters is essential in geotechnical projects and traditionally relies on laboratory tests, which can be expensive, time-consuming, and, in some cases, unfeasible. As a preliminary alternative, correlations based on in-situ tests can be used, however, such correlations are generally limited to sands or clays, restricting their applicability in many practical situations. A potentially more comprehensive solution is the development of models capable of relating the Standard Penetration Test (SPT), the most widely used field test in geotechnical practice, to the Mohr-Coulomb strength parameters, cohesion ( c ’) and friction angle ( ϕ ’), while accounting for different grain-size fractions. This task is challenging due to the high variability of the variables involved, however, by using machine learning techniques, such as artificial neural networks, which are capable of handling complex and nonlinear relationships, it has been possible to overcome part of these limitations. The proposed model also incorporates a relevant contribution: the numerical treatment of the “soil type” variable, allowing the network to account for differences among sandy, silty, and clayey soils. With a representative dataset, the model achieved correlation coefficients of 0.85 (training) and 0.90 (testing), with mean errors of 5.3 kPa for c ’ when c ’ < 20 kPa, and 2.36° for ϕ ’ in the range 25° ≤ ϕ ’ < 35°. In addition to demonstrating superior performance compared to models available in the literature, the model is capable of handling different grain-size fractions. It is concluded that the model shows satisfactory performance for preliminary applications in geotechnical engineering practice, allowing estimates of the strength parameters within well-defined validity ranges, however, it does not replace the need for laboratory testing.