In this paper we investigate how the KLM E&M WBBM Unit can improve the accuracy of non-routine tasks man-hours forecasts for the Wide-Body C-checks, per specific check visit and maintenance organization. Following a contingent approach, a database was set-up and three econometric forecasting methods are compared with the judgmental method currently used by the unit: linear regression analysis, artificial neural network and nearest neighbour analysis. Higher accuracy was obtained from the econometric methods when cross-validated, which proves that it is indeed possible to have a model to forecast non-routine maintenance work more accurately than judgmental methods.