This paper investigates the finite sample properties of confidence intervals for structural vector error correction models (SVECMs) with long-run identifying restrictions on the impulse response functions. The simulation study compares methods that are frequently used in applied SVECM studies including an interval based on the asymptotic distribution of impulse responses, a standard percentile (Efron) bootstrap interval, Hall's percentile and Hall's studentized bootstrap interval. Data generating processes are based on empirical SVECM studies and evaluation criteria include the empirical coverage, the average length and the sign implied by the interval. Our Monte Carlo evidence suggests that applied researchers have little to choose between the asymptotic and the Hall bootstrap intervals in SVECMs. In contrast, the Efron bootstrap interval may be less suitable for applied work as it is less informative about the sign of the underlying impulse response function and the computationally demanding studentized Hall interval is often outperformed by the other methods. Differences between methods are illustrated empirically by using a data set from King, Plosser, Stock & Watson (1991).