Bias errors for retrieved temperature profiles T(p) from the Halogen Occultation Experiment (HALOE) are evaluated by pairing with nearby soundings from ground-based Rayleigh lidar and from inflatable falling spheres. Findings for the mesosphere must be based on large sets of pairings because an individual HALOE T(p) measurement is somewhat noisy and may not view the same atmospheric structure. Simulated estimates of total bias errors for the HALOE profiles are of the order of 2% (or 5 K). The average biases with the lidar profiles at 44¿N, 6¿E are within 3 K from 33 to 66 km and 4 K to 74 km, at least when zonal easterlies (and reduced wave activity) prevail. Tidal effects can account for most of that difference. Comparisons with shipboard lidar soundings that begin near 100 km indicate no significant bias even in the 77 to 85 km range. Similar comparisons with falling sphere profiles at Cape Canaveral and Wallops Island indicate no significant bias from 40 to 66 km and 71 to 85 km. HALOE profiles are warmer than sphere values from 67 to 70 km, but this is also the altitude region where the sphere profile has greater uncertainties. For those months when the lidar soundings indicate inversion layers the sets of profiles paired with lidar indicate that the HALOE T(p) is somewhat colder in the upper mesosphere, in part owing to the effects of limb path averaging of horizontal structure by HALOE and its inability to completely resolve the inversion layer maximum. On the basis of the combined sets of findings it is concluded that most of the error in a single HALOE T(p) is random, except where there is a sharp inversion. Daily zonal averages of the HALOE profiles are accurate and useful for defining the seasonal and longer-term variations of T(p) in the mesosphere.