This paper proposes a hierarchic evolutionary model leading to long-ru
n cooperation in the Prisoner's Dilemma. The population is segmented i
nto groups, and evolution and interaction take place 'in parallel' at
two levels: (i) within groups at the lower level; (ii) among groups at
the higher one. At each of these levels, what performs currently best
tends to be imitated. On the other hand, occasionally, individuals al
so experiment with (or mutate to) some new strategy. If the number of
groups is sufficiently large, the long-run (invariant) distribution of
the process is shown to be concentrated on the cooperative outcome.