Infrared observations are used to obtain effective temperatures with a 3-percent error for a sample of low main-sequence stars. The empirical calibration of Arribas and Martinez Roger (1988) is found to be valid except for the lowest metallicities and for effective temperatures of lower than 5000 K, and to predict exceedingly blue colors for solar abundances. Better agreement with the observed behavior is found with the Buser and Kurucz (1988) model. The results support the claim of Magain (1987) that UV opacity is larger than that predicted using current atmospheric models for late-type stars.