Interannual climate variability impacts sugarcane yields. Local climate data such as daily rainfall, temperature and radiation were used to describe yields collected from three locations-Victoria sugar mill (1951-1999), Bundaberg averaged across all mills (1951-2010) and Condong sugar mill (1965-2013). Three regression methods, which have their own inbuilt variable selection process were investigated. These methods were (i) stepwise regression, (ii) regression trees and (iii) random forests. Although there was evidence of overlap, the variables that were considered most important for explaining yields by the stepwise regressions were not always consistent with the variables considered most important by the regression trees. The stepwise regression models for Bundaberg and Condong delivered a model that was difficult to explain biophysically, whereas the regression trees offered a much more intuitive and simpler model that explained similar levels of variation in yields to the stepwise regression method. The random forest approach, which extends on the regression tree algorithm generated a variable importance list which overcomes model sensitivities caused by sampling variability, thereby making it easier to identify important variables that explain yield. The variable importance list for Victoria indicated that maximum temperature (February-April), radiation (January-March) and rainfall (July-October) were important predictors for explaining yields. For Bundaberg, emphasis clearly centred on rainfall, particularly for the period January to April. Interestingly, the random forest model did not rate rainfall highly as a predictor for Condong. Here the model favoured radiation (February to April), minimum temperature (March-April) and maximum temperature (January to April). Improved understanding of influential climate variables will help improve regional yield forecasts and decisions that rely on accurate and timely yield forecasts.