Different farmers require different compensation payments to be incentivized to participate in water quality improvement-related agri-environmental schemes (AESs) and payment for ecosystem services (PESs). This is because they differ in their farm management practices, cost structures and attitudinal characteristics. However, these differences are rarely characterized in the design and implementation of AESs and PESs in the Global South. Using a discrete choice experiment, we investigate farmers’ willingness to accept compensation to control agricultural nonpoint source pollution in the Limpopo River Basin of South Africa to observe whether these differences matter. Conditional, random parameter and latent class logit models are estimated. Our latent class logit model identified one random choice class (farmers making random responses) and three preference classes of farmers (low-, moderate- and high-resistance) with dissimilar compensation requirements to alter their status quo farm management practices to improve water quality. Gender, age, education, farming experience and secure tenure rights are key drivers of preference heterogeneity.