Dairy data analysis may facilitate the selection of which cows should be culled to maintain or increase herd productivity. The objective of this study was to explore the usefulness of inductive learning methods (Decision Tree, DT) to identify cows with low productive expectations susceptible to voluntary culling. A database of 97,987 Holstein-Frisian cows provided by the Confederación Nacional de la Raza Frisona Española was used. The attribute Prod/dLife (average milk production/day of productive life, from first calving until culling) was selected to split the main root of the DT. The model input attributes, obtained from first lactation available information, were: peak milk data (kg milk/d, %fat, %protein and SCC from mo 2 and 3), reproductive indexes (calving-first AI interval and Days Open), cow’s birth season, age and season of first calving, genetic indexes of production (KL) and total merit (ICO), and morphological qualification. Numerical attributes were classified in 4 quartile classes: VL (very low), L (low), H (high) and VH (very high). Each DT branch represented a pattern met for those animals with the same classification for Prod/dLife and for other attributes. Patterns were considered significant when more than 10% of cows met the criteria. The most significant patterns were obtained with the attribute KgMilkPeak (average peak production of first lactation). From the 27,179 cows belonging to the quartile KgMilkPeak = VL, 64% were classified in the class Prod/dLife = VL and 22% in Prod/dLife = L. Therefore, 86% of the cows with a VL production at the first lactation peak had a VL or L average production in their productive life. This pattern could be used to classify cows according to milk yield of first lactation peak and to identify those with the lowest expected lifetime production. These results suggest that cows with poorly productive life could be detected in early stages of their first lactation and, thus, be classified as early culling candidates.