When reasoning about actions and sensors in realistic domains, the ability to cope with uncertainty often plays an essential role. Among the approaches dealing with uncertainty, the one by Bacchus, Halpern and Levesque, which uses the situation calculus, is perhaps the most expressive. However, there are still some open issues. For example, it remains unclear what an agent’s knowledge base would actually look like. The formalism also requires second-order logic to represent uncertain beliefs, yet a first-order representation clearly seems preferable. In this paper we show how these issues can be addressed by incorporating noisy sensors and actions into an existing logic of only-knowing.