The continuous growth in population and consumption, the intensity of competition for land, water and energy and the overexploitation of the ecosystem have all affected Africa's ability to sustain its food security and natural resources. In recent years, many promising agricultural development initiatives were unable to provide sustainable solutions to agricultural challenges in most parts of Africa, including Ghana, as a result of policy failures. The agricultural sector is a complex system and requires a holistic approach to deal with the root causes of challenges. This research therefore uses systems thinking tools, including causal loop diagrams and Bayesian belief network modelling, to develop new structural systems models whereby stakeholders can determine the components and interactions between the structure, conduct and performance (SCP) of the agricultural sector in Ghana, by using the first five steps of the Evolutionary Learning Laboratory. The results illustrate how the SCP elements interact together to influence the survival and growth of the agricultural sector. The study identifies that stakeholders adopt several strategies to survive and compete, which lead to overexploitation of the ecosystem. The results from the Bayesian belief network models indicate that the implementation of systemically determined interventions, policies and strategies could significantly improve the probability of business survival and growth from 58.8 to 73%. Also, the chances of improving the SCP could be increased from 39, 28.3 and 36.4 to 80.1, 55.9 and 62.4%, respectively, and these may vary based on the conditional probability tables. This paper contributes to the systemic approach to SCP, in that improvements to production and allocative efficiency may usher in a greater potential for improving food security, supporting the ecosystem and further strengthening agricultural sustainability. Copyright © 2016 John Wiley & Sons, Ltd.