We investigated the dependence of the geomagnetic activity index Kp on the velocity and density of the solar wind and the intensity of the interplanetary magnetic field (IMF). A three-layer neural network was used to create the model. The degree of the influence of input parameters on Kp was determined by the value of the mean and root-mean square deviations of the model index values from the real indices. It was found that the largest contribution to the Kp index is provided by the Z component of the IMF, the velocity and density of the solar wind measured with a delay from 0 to 3 h relative to the studied value of Kp, and the previous value of the index itself. For the model with such a set of input parameters, the correlation coefficient between model and real series is ±0.89. The analysis of deviations from the real values of Kp showed that high indices are simulated worse than low indices. In order to solve this problem the data distribution was reduced to a uniform distribution over Kp, and this considerably decreased the standard deviations for large values of Kp.