Animal experiments are widely required to comply with the 3Rs, to minimise harm to the animals and to serve certain purposes in order to be ethically acceptable. Recently, however, there has been a drift towards adding a so-called harm-benefit analysis as an additional requirement in assessing experiments. According to this, an experiment should only be allowed if there is a positive balance when the expected harm is weighed against the expected benefits. This paper aims to assess the added value of this requirement. Two models, the discourse model and the metric model, are presented. According to the former, the weighing of harms and benefits must be conducted by a committee in which different stakeholders engage in a dialogue. Research into how this works in practice, however, shows that in the absence of an explicit and clearly defined methodology, there are issues about transparency, consistency and fairness. According to the metric model, on the other hand, several dimensions of harms and benefits are defined beforehand and integrated in an explicit weighing scheme. This model, however, has the problem that it makes no real room for ethical deliberation of the sort committees undertake, and it has therefore been criticised for being too technocratic. Also, it is unclear who is to be held accountable for built-in ethical assumptions. Ultimately, we argue that the two models are not mutually exclusive and may be combined to make the most of their advantages while reducing the disadvantages of how harm-benefit analysis in typically undertaken.