Proper soil management techniques are essential to keep the soil healthy and without degradation. When this is not possible, this soil must be recovered, taking into account the attributes of the soil and its regenerative power, with this, several techniques are being used. In this context, this work aims to develop an interactive program (analyze and classify) using Artificial Neural Networks (ANN) to estimate soil recovery levels (recovered (R), partially recovered (PR) and not recovered (NR) as a function of physical attributes. The experiment was carried out at the São Paulo Agribusiness Technology Agency - APTA do Extremo Oeste, in Andradina / SP from 2015 to 2017, in soil classified as Ultisol cultivated with Urochloa pasture, with different ways of introducing Estilosantes cv. Campo Grande. The soil attributes studied were: soil density, soil porosity, mechanical resistance to penetration, water infiltration in the soil and weighted average diameter in the soil layers: 0-10; 0.10-0.20 and 0.20-0.40 m The program was developed in the MATLAB environment and the simulation was performed using a graphical interface. and work was the multilayer Perceptron (MLP). It was found that the network achieved adequate training, with a low mean square error, which could generate an interesting and automatic alternative for the classification and analysis of recovering soils. The results were printed on a self-explanatory graphical interface, with graphs and metadata of the physical indexes and their classifications regarding ANN.