In the evaluation of pedotransfer functions (PTF) for estimating soil water retention (SWR), not much attention is given to error pattern. However, its relation to soil properties can reveal important details, thus increasing PTF reliability. Decision tree models constitute a promising alternative for the identification of these characteristics because of its interpretable and symbolic representations. The objective of this study was to evaluate the error patterns of the PTF developed by Vereecken and his collaborators in 1989 (PTFv), by means of decision tree induction. The evaluation was performed for SWR at FC (-9.8 kPa) and at PWP (-1,500 kPa) on the same data set used in the generation of the PTFv. Sand (Sa), silt (Si) and clay (C) content, bulk density (BD) and carbon organic content (CO) were used as predictors. The occurrence (“Error”) or the non-occurrence (“NonError”) of error in the estimations was considered for the target attribute. At FC, 19% of the values were classified as “Error” and 81% as “NonError”. At PWP, 32% of the values were classified as “Error” and 68% as “NonError”. The models created were simple, with 91% accuracy for FC and 70% for PWP. At FC, the main rule derived was that for horizons with C ≤ 3.5% and OC ≤ 0.19%, the class is “Error”. For the class “NonError”, the main rule comprised soils with C > 3.5%. At PWP, soils with Sa ≤ 33.9%, OC ≤ 1.27% and BD ≤ 1.5 gcm^-3, the estimates were erroneous. Sa > 33.9% and Si > 4.5% represented one of the rules obtained for the class “NonError”. The decision tree technique seemed to be efficient to evaluate error pattern of PTFv. Despite the low number of errors in the estimations, the technique made it possible to identify the conditions under which they occurred.