This paper investigates strategies for real-time estimation of the output gap. First, I examine estimates from univariate models with stochastic cycles. This corresponds to the use of model-based band-pass filters in real-time, and I find that the turning points in real-time and final output gap series match more closely for higher order models and that the revisions properties and real-time accuracy are more favorable. Second, I investigate the use of capacity utilization as an auxiliary indicator to improve on output gap estimates in real-time. I find that this bivariate approach leads to significant gains in the accuracy of real-time estimates and in the quality of revisions.