ABSTRACT Brazil is the largest Coffea arabica L. producer and exporter in the world market. In view of the need for nitrogen fertilization, the objective of this work was to evaluate the use of digital images of coffee leaves as a function of nitrogen concentration and cultivars, as well as to evaluate the most adequate sampling of this technique to predict leaf nitrogen. The experiment was set in a greenhouse, at UNESP - Experimental Campus of Registro-SP, using 12L pots with medium sand. A completely randomized design was adopted, in a 4 x 2 factorial scheme, with four concentrations of N (0, 50, 100 and 200 mg dm-3), and two coffee cultivars (Mundo Novo and Obatã), with ten replications. The variables evaluated were: hue, dark green color index (DGCI), leaf area and leaf nitrogen content. The hue and DGCI were influenced by N concentration, cultivar and are dependent on the day after treatment application and the number of leaves used. The use of five leaves per replication was more adequate to estimate leaf N content. Hue, DGCI and leaf area were higher in Obatã depending on the day after application of the treatment and N concentration.