For many years, Load Forecasting (LF) has been an area of intense research. Most research has focused on short and long-term forecasting, with "short-term" generally meaning one hour in advance, to enable some power grid operations tasks. However, finer-grained prediction, at a resolution of minutes, can assist with other tasks, such as Power System State Estimation, and matching load to renewable energy generation in developing Smart Grids. To allow such ephemeral-term prediction with high accuracy, analysis of historical data sampled at high frequency is necessary. In this paper, we present statistical analysis based on three years' worth of real data obtained from the New York Independent System Operator (NYISO). The data is fine-grained, at a resolution of one sample per five minutes. The advantage of this data set is the ability to verify the applicability of our results to both large and small systems. The data and analysis presented in this paper can be used as a baseline for future LF and Smart Grid research.