The on-farm benefits of agricultural research or extension often depend in complex ways on the way that the new technology or information affects the farming system. The existence of farming system-wide impacts of research and extension is highlighted as a critical but neglected issue in economic evaluations of agricultural research. The issue is neglected in the available texts on agricultural research evaluation and in most applied evaluations. Given the difficulty and complexity of accurate benefit estimation, we see a renewed role for farm-level economic models (such as whole-farm linear programming models) in this area. The benefits of undertaking a more sophisticated and detailed analysis to estimate research benefits include not just greater accuracy but also greater credibility with researchers and greater relevance through representing factors which they perceive to be important. The paper discusses how, if such respect is engendered, a formal research evaluation can yield additional benefits by improving the design of research. (C) 1999 Elsevier Science Ltd. All rights reserved.