This research is focused on validating the performance of a simple geostatistical interpolation method to produce predictions at certain points, in case no predictions or data are available at these points. A two-step procedure to perform the predictions at certain stations using Inverse Distance Weighting (IDW) has been developed. On the one hand, the first stage is to select the best Artificial Neural Network (ANN) prediction model for each monitoring station, and then, predictions are performed for all stations with these models. Then, in the second stage, we used the Leave-One-Out (LOO) validation method together with IDW, leaving one station out at a time, to validate how the method operates and to be able to produce the predictions at a point such as a pollution monitoring station (where predictions are usually performed with ANNs). The approach was applied to PM10 air pollutant in order to check its performance. This IDW method proves to be effective, obtaining a mean correlation coefficient (R) value of 0.7355 and an average Mean Squared Error (MSE) of 733.74, which are very similar but slightly lower than those obtained directly with ANNs (mean R of 0.8541 and MSE of 726.53) using a resampling procedure. The method could be useful when historical data from a station is not available and ANNs cannot be used to make predictions, thus allowing the use of predictions from other stations and the IDW method.