On upland soils in tropical Africa, common production constraints of rice and maize on smallholder farms are poor soil fertility—resulting from soil erosion and nutrient depletion—and infestation by witchweeds (Striga spp.). In Madagascar where these crops are often grown in rotation, combining legume cover crops with no-till and crop residue mulching—labelled conservation agriculture (CA)—may address these problems. Previously, it was shown that CA practices contribute to steep reductions in Striga asiatica infection. In the current study, a 4-year field experiment was conducted to test, for the first time, the hypothesis that CA practices also contribute to crop yield and soil improvements under Striga-infested conditions. The conventional mono-crop rice–maize rotation practice, involving seasonal tillage and crop residue removal, was compared to three rice–maize rotation systems following CA practices, each with a different legume cover crop option: (1) two short-cycle annual species, cowpea (Vigna unguiculata) and mucuna (Mucuna pruriens); (2) a long-cycle annual, ricebean (Vigna umbellata); and (3) a perennial, stylosanthes (S. guianensis). Rice yields, as well as yield variability, generally increased by changing from the conventional to a CA practice, and maize yields were variable and low in particular under the CA practices. CA practices significantly reduced soil displacement by rainwater runoff and increased soil nitrogen and pH levels (0–20 cm depth), in particular with stylosanthes as cover crop, but did not result in a significant change in soil organic carbon concentration. Rice yields correlated negatively with Striga asiatica plant numbers in years with moderate infection levels. This is the first study that shows mixed outcomes from CA practices in tropical cereal rotation systems on degraded, Striga-infested soils, and subsequent entry points for system improvements. Suggested improvements include judicious cover crop management, complementary fer