We present in this work a novel methodology for the co-clustering of functional data within a probabilistic framework. For this, following the model-- based approach of G. Govaert and M. Nadif for continuous or categorical data, we propose a latent functional block model. This model relies on the following assumptions: 1. conditionally on the block the data belong, they are assumed inde- pendent and identically distributed. 2. the column and row partition are assumed to be independent. 3. within a block, the data are modeled by a parsimonious high- -dimensional Gaussian model onto the expansion coefficients of the curves into an appropriate finite basis of functions.