Using similar states in a database called “catalogue” such as a reanalysis, analogues provide simple yet efficient ensemble forecasts in atmospheric and ocean sciences. Typically performed on low-resolution images of large-scale atmospheric or ocean circulation, analogue forecasting encounters uncertainties due to unresolved small spatial scales, as the latter contribute to the time-evolution of the circulation but not to the similarity criterion used to search for analogues. Another source of uncertainty are the finite distances between the analogues and the initial target large-scale variables. We disentangle these two sources of uncertainty using a modified version of the Lorenz system, where stochastic terms account for unresolved small spatial scales. For large enough catalogue size and forecast horizon, we show that the analogue forecasting ensemble spread is dominated by the effect of stochastic terms, with only little influence of the initial analogue-to-target distances. Conversely, for short-term forecast and small catalogue size, the analogue ensemble is mostly influenced by initial analogue-to-target distances and not by the effects of unresolved scales. This result calls for adjustments of the classical analogue method for small forecast horizons.