In this paper, we adapt to cell-free Massive MIMO (multiple-input multiple-output) the finite-blocklength framework introduced by \"Ostman et al. (2020) for the characterization of the packet error probability achievable with Massive MIMO, in the ultra-reliable low-latency communications (URLLC) regime. The framework considered in this paper encompasses a cell-free architecture with imperfect channel-state information, and arbitrary linear signal processing performed at a central-processing unit connected to the access points via fronthaul links. By means of numerical simulations, we show that, to achieve the high reliability requirements in URLLC, MMSE signal processing must be used. Comparisons are also made with both small-cell and Massive MIMO cellular networks. Both require a much larger number of antennas to achieve comparable performance to cell-free Massive MIMO.