Aims: We estimate the minimum length on which solar granulation can be considered to be a Markovian process. Methods: We measure the variation in the bright difference between two pixels in images of the solar granulation for different distances between the pixels. This scale-dependent data is empirically analyzed to find the minimum scale on which the process can be considered Markovian. Results: The results suggest that the solar granulation can be considered to be a Markovian process on scales longer than r_M=300-500 km. On longer length scales, solar images can be considered to be a Markovian stochastic process that consists of structures of size r_M. Smaller structures exhibit correlations on many scales simultaneously yet cannot be described by a hierarchical cascade in scales. An analysis of the longitudinal magnetic-flux density indicates that it cannot be a Markov process on any scale. Conclusions: The results presented in this paper constitute a stringent test for the realism of numerical magneto-hydrodynamical simulations of solar magneto-convection. In future exhaustive analyse, the non-Markovian properties of the magnetic flux density on all analyzed scales might help us to understand the physical mechanism generating the field that we detect in the solar surface.