In the past few years various methods of identifying structural dynamics models from modal testing data have appeared. This paper presents a comparison of four of these algorithms: the Eigensystem Realization Algorithm (ERA), the modified version ERA/DC, where DC indicates that it makes use of data correlations, the Q-Markov cover algorithm, and an algorithm due to Moonen, DeMoor, Vandenberghe and Vandewalle (MDVV). The comparison is made using a five-mode computer model of the 20 m Mini-Mast truss structure at the NASA Langley Research Center, and various noise levels are superimposed to produce simulated data. The results show that, for the example considered, ERA/DC generally gives the best results; that ERA/DC is always at least as good as ERA, which is shown to be a special case of ERA/DC; that Q-Markov requires the use of significantly more data than ERA/DC to produce comparable results; and that in some situations Q-Markov cannot produce comparable results. The MDVV algorithm can avoid pre-processing of data to obtain Markov parameters, but requires significantly larger matrix singular value decomposition for large data sets.