Brazil is the world’s sixth largest cocoa producer with 270,000 tonnes of cocoa produced per year. In a world with an increasing demand for chocolate, but with agriculture threatened by climate change, the chocolate industry is worried about a possible shortage of cocoa. Furthermore, growing cocoa is a main cause of deforestation. However, in the state of Bahia, Northeast Brazil, cocoa is grown in traditional agroforests called ‘cabrucas’ which maintain a forest cover. Cocoa, an understorey crop, is planted under the shade of native Atlantic Forest trees and exotic fruit trees introduced by the farmers. These cabrucas have high conservation value but very low cocoa yield. In my thesis I investigate the factors limiting cocoa yield and how to increase yield in cabrucas. I explore the role of shade trees and the nutrient dynamics in litterfall. Finally, I explore the risk that climate change could represent for cocoa production in the future by looking at the effects of an unexpected drought caused by an El Niño Southern Oscillation (ENSO) event between November 2015 and May 2016. My study is based on data collected in permanent transects in 32 randomly chosen cabruca farms in Barro Preto a municipality of Bahia over a period of three years. I also established littertraps in 10 cocoa farms where I intensively studied nutrient dynamics and cocoa yield per tree over 12 months. My results showed that unproductive trees, low cocoa tree density, high shade cover and high cocoa mortality due to drought were the main factors limiting cocoa yield in cabrucas. Surprisingly, adding fertilisers to the cocoa trees did not increase yield. This suggests that there is no nutrient deficiency in cabrucas. In the farms, I found 69 species of shade trees for an average density of 125 ± 32 trees per hectare. Half of the species of shade trees were Atlantic Forest species of conservation value. The litterfall experiment showed the shade tree species and the quantity of litterfall produced, can affect the number of cocoa pods per tree. In cabrucas, a higher number of cocoa pods was found on cocoa trees under shade trees than under no shade. Finally, I showed that the exceptionally severe ENSO-related drought caused 80% loss in yield and 11% cocoa tree mortality in Barro Preto. Climate models predict an increased frequency of strong ENSO events in the future. Farmers in Bahia are not prepared to face regular drought events. The 2015/16 drought affected the dynamics of cocoa production in Brazil: it accelerated the decrease of extensive wildlife-friendly cocoa production in Bahia whereas it increased the development of cocoa production in intensive low shade plantations in the state of Pará. This suggests that climate change could be a threat to traditional cocoa agroforests in Bahia. Developing wildlife-friendly certification schemes and Payment for Ecosystem Services to internalise the value of forest conservation and to encourage farmers to maintain their shade trees could save cabruca systems from going extinct.