Three time-series models (regression, decomposition, and Box-Jenkins autoregressive integrated moving averages) were applied to national surveillance data for campylobacteriosis with the goal of disease forecasting in three US states. Datasets spanned 1998-2007 for Minnesota and Oregon, and 1999-2007 for Georgia. Year 2008 was used to validate model results. Mean absolute percent error, mean square error and coefficient of determination (R2) were the main evaluation fit statistics. Results showed that decomposition best captured the temporal patterns in disease risk. Training dataset R2 values were 72.2%, 76.3% and 89.9% and validation year R2 values were 66.2%, 52.6% and 79.9% respectively for Georgia, Oregon and Minnesota. All three techniques could be utilized to predict monthly risk of infection for Campylobacter sp. However, the decomposition model provided the fastest, most accurate, user-friendly method. Use of this model can assist public health personnel in predicting epidemics and developing disease intervention strategies.