The monthly pattern of suicides has remained a puzzle ever since it was discovered in the second half of the 19th century. In this paper we intend to "explain" not the pattern itself but rather its changes across countries and in the course of time. First, we show that the fairly common idea according to which this pattern is decaying in "modern" societies is not altogether true. For instance, around 2000, in well urbanized countries like South Korea or Spain this pattern was still as strong as it was in France (and other European countries) in the late 19th century. The method that we use in order to make some progress in our understanding is the time-honoured Cartesian approach of breaking up the problem under consideration "into as many parts as might be necessary to solve it". More specifically, we try two decompositions of monthly suicides: (i) according to suicide methods (ii) according to age-groups. The first decomposition points out the key-role of hanging and drowning. The second shows the crucial role of the $ 15-20 $ and $ 65+ $ age-groups. Then, we present a number of cases in which age-group decomposition provides adequate predictions. It turns out that the cases in which the predictions do not work are newly urbanized countries. The discrepancies may be due to a memory effect which induces a time-lag extending over one or two generations. Finally, in the light of the new results presented in the present paper, we re-examine the theory proposed by Emile Durkheim.