Risky research projects are, other things being equal, intrinsically harder to monitor than projects that are less risky. It is shown using agency theory that a standard cost benefit analysis, which ignores the agency problem, will introduce a bias towards excessively risky projects, and it will under‐estimate the benefits from complementary investments in libraries, scientific equipment and other expenditures that increase the productivity of scientists. Research managers should be risk‐averse in their choice of projects, and they should aim to hold a balanced portfolio of projects. The nature of this portfolio problem is, however, quite different from the portfolio management problem in financial markets.