Accurate, real time air quality measurements are difficult to make, because real time sensors for some gas species are not specific to a single gas. For example, some carbon dioxide sensors react to hydrogen sulfide. By combining the response of several types of real time gas sensors the Real-time Air Quality Monitoring System (RAQMS) accurately measures many different gases. The sensor suite for the INEEL's Real-time Air Quality Monitoring System (RAQMS) incudes seven, inexpensive, commercially-available chemical sensors for gases associated with air quality. These chemical sensors are marketed as devices to measure carbon dioxide, hydrogen sulfide, carbon monoxide, sulfur dioxide, nitrogen dioxide, water vapor and volatile organic compounds (VOC's). However, these chemical sensors respond to more than a single compound, e.g. both the VOC and the carbon dioxide sensors respond strongly to methane. This multiple sensor response to a given chemical is used to advantage in the RAQMS system, as patterns of responses by the sensors were found to be unique and distinguishable for several chemicals. Therefore, there is the potential that the seven sensors combined output can: (1) provide more accurate measurements of the advertized gases and (2) estimate the presence and quantity of additional gases. The patterns of sensor response can be thought of as clusters of data points in a seven dimensional space. One dimension for each sensor's output. For all of the gases tested, these clusters were separated enough that good quantitative results were obtained. As an example, the prototype RAQMS is able to distinguish methane from butane and predict accurate concentrations of both gases. A mathematical technique for estimating probability density functions from random samples is used to distinguish the data clusters from each other and to make gas concentration estimates. Bayes optimal estimates of gas concentration are calculated using the probability density function. The Bayes optimal estimates are analogous to least square error curve fits or regression analysis. A computer program was used to find the best parameters for the Bayes optimal estimating functions. The program implemented a probabilistic neural net.