Traditional reasoning on statistical effects in energy loss and multiple scattering phenomena in random media are based on the concepts of binary interactions, full independence of successive events and finite variance of random energy transfers or angular deflections in single encounters. These assumptions imply Poissonian statistics and a rapid convergence of the corresponding distributions towards the Gausssian with r.m.s.∼( depth) 1 2 ; moreover they may hold rigorously in dilute media only. We shall examine various effects which may appear when the above constraints are relinquished. Thus, density effects in condensed media and the recently evidenced yield enhancement in 180° RBS are due to a breakdown of full independence; the yield enhancement for thick targets near narrow resonances termed the Lewis effect is due to strictly non-Gaussian behaviour of small energy losses. As for infinite variances, they play a major role in multiple scattering or in electron energy loss processes leading to a complete breakdown of the usual ( depth) 1 2 rule of distribution broadening. The key-role of the theory of infinitely divisible and of stable probability densities in particle-solid interaction theory will be discussed.