Cascade impactors are used to study the in vitro deposition of aerosol products and are accepted worldwide by regulatory bodies, such as the United States Federal Drug Administration, as a means of comparing the bioequivalence between products. They are also employed in general research and development activities. The method of calculating aerosol parameters from measurements in these apparatus is prone to operator error and collection of the drug deposited can be time consuming. The purpose of this study is to determine the feasibility of using artificial neural networks (ANN) to predict aerosol deposition using different formulations and device characteristics. Published data1 was used to train and test the ANN using fine particle fraction (%) as the dependent variable. Unscrambler™ was used to minimise the number of descriptors in the input vector, by removing fields that did not contribute significantly to explaining the variance in the dataset. The significant descriptors were combined to form an input vector that was used to search the predefined ANNs in the Neurosolution software™ and to identify the optimum architecture. A multilayer perceptron proved the best architecture for this data set by producing high R2 value for training (0.99), cross validation (0.99) and test set (0.96) and low errors between the desired output and the output generated from the ANN. In conclusion, this pilot study showed that ANN is a promising technique, and could be readily applied to large, diverse datasets to generate a robust and predictive model capable of predicting a cascade impactor output.