Using animals in scientific research is commonly justified on the utilitarian basis that the benefits of scientific progress to human health and society exceed by far the harm inflicted on animals. In an attempt to ensure that this is indeed the case for every research project, legislation and guidelines increasingly demand the application of harm-benefit analysis (HBA) as part of the approval process of animal research protocols. The ethical principle of HBA asserts that the costs of an action should be weighed against the expected benefits. Any action that may inflict harm can only be approved if it is associated with a greater benefit. This principle is intuitively appealing but how to use it as a practical rule for ethical decisions is a difficult question. The main difficulty is that the future benefits of most scientific research are unmeasurable, unpredictable and are not manifested at the level of the individual project. Applying HBA in such cases may impede scientific progress by inducing a bias against basic research. Moreover, it can lead to the toleration of unnecessary harm to animals in research. Given these caveats of HBA, I call policy-makers to reconsider the place of HBA in animal research. Instead, I support an alternative guideline which is based on replacing the HBA principle (that the expected benefits of the research must exceed the harms caused to the animals) with two independent but mutually necessary principles: (1) any research using an animal must carry a benefit for society and (2) the harm inflicted to an animal in an experiment must be minimal and scientifically justified. I argue that rigorous harm-analysis, which is not weighted against obscure benefits, can increase the over-all benefits of research while reducing the harms to animals.