The typical approach of estimating crop response to future climate scenarios may be inappropriate in the case of smallholder multi cropping rain-fed agriculture. Indeed, a crop-by-crop simulation, based on current varieties, cannot take into account the dynamics among crops as well as within crops i.e. among varieties, in time and space. We implemented a comparative study to understand interactions between cropping system dynamics and pas climate variations, taking into account the diversity of farmers' experiences and socio-cultural organization. in Kenya, farmers who adopted maize a few years ago are still cultivating traditional sorghum and pearl millet varieties, while others abandoned them earlier in favour of maize. Farming systems were thus dynamic, with different crop assemblages over time. Thus, retrospectively, farmers' capacity to mitigate crop failure risk due to extreme rainfall events has never been constant. Has the farming system lost part of its capacity to cope with climate variability, as maize is known to be less resistant to drought than sorghum and pearl millet? While this is usually demonstrated using yield parameter, we used see losses, which is consistent with a multicrop system. Combining ecological anthropology and climatology, we confronted the results of a retrospective survey of farmers' seed loss reminiscence about the period 1961-2006 and climatic records for three altitudinal levels on the eastern slope of Mount Kenya were analysed. Over the period 3204 seed loss events were reported independently by 208 farmers, for eight main crops of their rain-fed farming systems. the causes given for these losses according to farmers' experience and knowledge were recorded yearly. We first assessed whether these causes were related to recorded rainfall values, and, second, analysed the proportion of lost seed on a yearly basis, crop by crop and on the whole farming system, using logistic regression. Drought was mentioned 73.5% of the time wherea