The motivation for this paper is to investigate the efficiency of a Neural Network (NN) architecture, the Psi Sigma Network (PSN), in forecasting US unemployment and compare the utility of Kalman Filter and Support Vector Regression (SVR) in combining NN forecasts. An Autoregressive Moving Average model (ARMA) and two different NN architectures, a Multi-Layer Perceptron (MLP) and a Recurrent Network (RNN), are used as benchmarks. The statistical performance of our models is estimated throughout the period of 1972-2012, using the last seven years for out-of-sample testing. The results show that the PSN statistically outperforms all models’ individual performances. Both forecast combination approaches improve the statistical accuracy, but SVR outperforms substantially the Kalman Filter.