Ground‐based observations from three middle‐ and high‐latitude sites managed by the U.S. Department of Energy's Atmospheric Radiation Measurement (ARM) program are used to determine the sensitivity of the low‐cloud optical depth to temperature and to test whether observations support mechanisms previously proposed to affect the optical depth feedback. Analysis of cloud optical depth retrievals supports previous satellite findings that the optical depth decreases or stays constant with increases in temperature when the cloud is warm but increases when the cloud is cold. The cloud liquid water path sensitivity to warming largely explains the optical depth sensitivity at all sites. Mechanisms examined in this study involve the temperature dependence of (a) the moist‐adiabatic lapse rate, (b) cloud phase partitioning, (c) drying efficiency of cloud top mixing, (d) cloud top inversion strength, and (e) boundary layer decoupling. Mechanism (a) is present across all clouds and explains 30% to 50% of the increase in liquid water path with warming at temperatures below 0 °C. However, the cloud's adiabaticity, the ratio between the liquid water path and its theoretical maximum, is at least as important and determines how the liquid water path sensitivity to temperature varies with temperature. At temperatures below 0 °C, the adiabaticity increases with warming, and the data support mechanism (b). At warmer temperatures, the adiabaticity decreases with warming, overwhelming mechanism (a) and resulting in the liquid water path decreasing with warming. This adiabaticity decrease arises primarily because of mechanism (d), and to a lesser degree because of mechanism (e). No evidence is found supporting mechanism (c).