Examining a sample of massive galaxies at 1:4 < z < 2:5 with K Vega < 22 from GOODS, we compare photometry from Spitzer at mid-and far-IR to submillimeter, radio, and rest-frame UV wavelengths, to test the agreement between different tracers of star formation rates (SFRs) and to explore the implications for galaxy assembly. For z $ 2 galaxies with moderate luminosities (L 8 m < 10 11 L ), we find that the SFR can be estimated consistently from the multiwavelength data based on local luminosity correlations. However, 20%Y30% of massive galaxies, and nearly all those with L 8 m > 10 11 L , show a mid-IR excess that is likely due to the presence of obscured active nuclei, as shown in a companion paper. There is a tight and roughly linear correlation between stellar mass and SFR for 24 mYdetected galaxies. For a given mass, the SFR at z ¼ 2 was larger by a factor of $4 and $30 relative to that in star-forming galaxies at z ¼ 1 and 0, respectively. Typical ultraluminous infrared galaxies (ULIRGs) at z ¼ 2 are relatively ''transparent'' to ultraviolet light, and their activity is long lived (k400 Myr), unlike that in local ULIRGs and high-redshift submillimeter-selected galaxies. ULIRGs are the common mode of star formation in massive galaxies at z ¼ 2, and the high duty cycle suggests that major mergers are not the dominant trigger for this activity. Current galaxy formation models underpredict the normalization of the mass-SFR correlation by about a factor of 4 and the space density of ULIRGs by an order of magnitude but give better agreement for z > 1:4 quiescent galaxies.