As both the natural level of output and the New Keynesian output gap are unobserved, there is quite some debate on the question how these variables look like in practice. Rather than taking the standard approach of using a time trend or the HP-filter to obtain estimates of these two objects, this paper takes a theoretically more sound route by separating trend from cycle via Bayesian estimation of a New Keynesian model, augmented with an unobserved components model for output. This delivers us with model consistent estimates of both the natural level of output and the New Keynesian output gap. These estimates are then compared with the dominant output gap proxies used in the literature. It turns out that the benefits of using the model-based approach taken in this paper mainly emerge in real time, thereby making this method potentially useful for the conduct of monetary policy.