Particle motion characteristics of short-period three-component (3-C) data are compared for various seismic phases at NORESS and ARCESS, and their usefulness for phase identification is evaluated. Continuous recordings at the arrays of 3-C elements were processed during routine operation of the Intelligent Monitoring System (IMS). The data set used in this study consists of 3822 arrivals extracted from the IMS database and covers a period of about 2.5 months. First, polarization attributes and azimuth of the dominant linear motion are compared for local/regional phases (Pn, Pg, Sn, Lg) at the two arrays. P-type arrivals have larger angles of incidence at ARCESS than at NORESS, on average, for similar ranges of distance and signal-to-noise ratio (SNR). This can be partly explained by higher crustal velocities under the ARCESS array. Also, at ARCESS the ratio of horizontal to vertical power is similar for Sn and Lg, on average, while at NORESS it is larger for Sn, Sn and Lg azimuths at ARCESS (and Lg azimuth at NORESS, with more scatter) provide good estimates of backazimuth (with a 180° ambiguity) and indicate predominance of SH motion at ARCESS. In the second part of this study, multivariate data analysis is performed to obtain phase identifications (with associated confidence), using polarization attributes as predictors. P- and S-type phases are distinguished with a success rate of 82% at NORESS and 89% at ARCESS. The performance is even better for P-type arrivals with 3-C SNR > 2 (96 and 98%, respectively). Sn and Lg are correctly identified for 74% of S-type phases at NORESS and 64% at ARCESS. Contamination of Lg by Sn coda at shorter ranges and dominant SH motion for both S-type phases at ARCESS affect the performance. This study shows the importance of evaluating each 3-C station individually.