A precise model of flowering at both the individual and the orchard level is an important prerequisite in determining final production for fruit species with particular pollination requirement such as the strawberry guava (Psidium cattleianum). Most flowering models have been designed for orchards, implicitly assuming that all the trees have the same flowering pattern. Their accuracy may be affected by flowering variability between trees, linked to genetic and/or environmental factors. A stochastic, temperature driven, flowering model developed for kiwi orchard was tested on strawberry guava. This model had two parameters: a flowering threshold a and a parameter of variance relative to flowerind sigmax2. Data were obtained from the flowering of 3 sets of trees, all genetically different, located at 200, 520 and 890 in of elevation in Réunion Island. Data collected at 520 in were used to fit the model, while data obtained at 200 and 890 in were used to validate it. The model described well the flowering at 520 in, but failed to account for flowering at 200 and 890 in and to represent the individual flowering patterns on each site. Analysis of discrepancies between model and data showed that the variability between trees should be considered: number of flower per tree, time and duration of individual flowering. The previous stochastic model has thus been re-estimated at the tree level and the over-all flowering was considered as the superposition of the individual flowerings weighted by their respective number of flower. It had then 2P (P = number of trees) parameters. It described well the individual flowering pattern, and represented more accurately the over-all flowering per site. A more general model needs however to be derived from it, while reducing the number of parameters.