Short-term electricity load forecasts and long-term projections of climate change impacts can benefit from understanding the relationship between electricity demand and meteorological conditions. We developed and applied a segmented regression technique to more than ten years of hourly electricity load data to estimate this relationship in two transmission zones in the United States that vary in their spatial extent and population. We empirically determined reference temperatures for cooling- and heating-degree hours. These reference temperatures differ from each other for every hour of the day and vary in accordance with the ambient temperature, which affect electricity loads induced for heating and cooling. Past temperatures and relative humidity have a significant influence on electricity load, and we identified the existence of threshold temperatures for the effect of relative humidity. Our results suggest that accurate predictions of the electricity loads should incorporate a ∼7 °C “comfort zone” where electricity load is less sensitive to temperature than elsewhere in the relationship, include the dependence on relative humidity (which can be negative), and incorporate a path dependence of prior days' temperatures.