During the last decade, the city of Austin has witnessed two important cold snaps, one of which has brought the system load very close to peaking during otherwise average winter months. We examine the effect of a cold snap on load curves. To do so, we have estimated a time-series hourly load model allowing for both time-dependent and weather-dependent activities, and for serial correlation. We find three significant results. First, there is a significant first-order serial correlation, and this autoregressive process must be accounted for. Otherwise strong doubts are cast upon results based on models which do not. Second, not only is the system-load consumption significantly affected by unusual and extreme weather conditions, but there is also marked difference between load curves for an average and a peak winter day. Finally, the severity of the cold snap has a stronger influence on system load in both increasing average consumption levels and modifying the shape of the load curves than does the length of the cold snap itself. Given the important impact of changes in outside temperature on system load in general and the (sometimes heated) debates surrounding the construction of new generating and distributing facilities, these findings should be of general interest.