Trophic cascades are important determinants of food web dynamics and functioning, yet mechanisms explaining variations in trophic cascade strength remain elusive. Here we combine a mesocosm experiment (phytoplankton-zooplankton-shrimp) and theoretical models to disentangle the relative importance of two processes driving trophic cascade: primary productivity (productivity hypothesis) and energy transfer efficiency (energy transfer hypothesis). Our experimental systems showed consistently positive effects of trophic cascade across mesocosms with different nutrient inputs and zooplankton communities. Moreover, the strength of trophic cascade increased with the energy transfer efficiency between herbivores and predators, but did not change with the primary productivity. These findings are further corroborated by our analyses of food chain models in more general settings. Combined, our results support the energy transfer hypothesis but falsify the productivity hypothesis. Our study contributes an integrative perspective to reconcile energetic and population dynamics in food webs, which has useful implications for ecosystem management.