The presence of cemented soils pose significant challenges in drilled shaft design and may prevent accurate estimates of the service limit state if traditional analytical techniques are employed. Thus, an Artificial Neural Network (ANN) is developed and tested as an alternative method for predicting settlement induced by axial loads. Training is carried out using the results of 31 field load tests performed in Las Vegas, USA, where cemented soils are common, and an automated process is employed to determine the optimal network architecture. Ultimately, a cascaded feed-forward ANN with one hidden layer consisting of six artificial neurons produced the highest quality generalization. Ten additional load tests not included in the original training, testing, or validation datasets are reserved to evaluate performance. It is observed that the ANN produces similarly accurate estimates of load-settlement on average as compared to two more traditional t-z style approaches.