The effectiveness of the perennial plants breeding is restricted by particular constraints linked to the crops lifespan. Moreover, acting of a production of fruits or seeds as to the coffee-tree, how many years of observation are necessary to define the trait of "production"? And how to integrate the variations of the trait in time with a breeding purpose in which stability and durability are often mentioned? To try to answer these questions, a trial comparing 20 clones of Coffea canephora, monitored over 9 years of production, was analyzed according to various methods. Two main methods were proposed: i) the first one proposed indexes definition - the indexes described precocity, alternation and intensity of inter-annual variations; ii) the second one proposed a modelization of the data by a longitudinal data analysis approach. It was found that the first cycle (first 4 years) are not always sufficient for predicting the yield capacity of the clones. The intensity of the relative differences between yields in successive years was a trait that was only heritable in the second production cycle. Low intensities for those differences were favorable to cumulative production. Strong alternations were not propitious to optimizing cumulative production. Several models were tested for longitudinal data analyses and Compound Symetric's one with heterogeneous variances (CSH model) allowed to better describe data structure. Thus, correlations between yields of successive years were pretty stable that reveals an important trees' effect among this clonal population. Variability was found for yield distribution over time. It is therefore possible to define ¿stability¿ traits and to select cultivars with lower annual yield variations, thereby ensuring regular income for coffee farmers.