Challenges that smart grids aim to address include the increasing fraction of supply by renewable energy sources, as well as plain rise of demand, e.g., by increased electrification of transportation. Part of the solution to these challenges lies in exploiting the opportunity to steer residential electricity consumption (e.g., for flattening the peak load or balancing the supply and demand in presence of the renewable energy production). To optimally exploit this opportunity, it is crucial to have insights on how flexible the residential demand is. Load flexibility is characterized by the amount of power, time of availability and duration of deferrable consumption. Residential flexibility however, is challenging to exploit due to the variation in types of customer loads and differences in appliance usage habits from one household to the other. Existing analyses of individual customer flexibility behavior in terms of timing are often based on inferences from surveys or customer load patterns (e.g., as observed through smart meter data): there is a high level of uncertainty about customer habits in offering the flexibility. Even though some of these studies rely on real world data, only few of them have quantitative data on actual flexible appliance usage, and none of them characterizes individual user behavior. In this paper, we address this gap and contribute with: (1) a new quantitative specification of flexibility, (2) two systematic methodologies for modeling individual customer behavior, (3) evaluation of the proposed models in terms of how accurately the data they generate corresponds with real world customer behavior, and (4) a basic analysis of factors influencing the flexibility behavior based on statistical tests. Experimental results for (2)-(4) are based on a unique data set from a real-life field trial. (C) 2016 Elsevier Ltd. All rights reserved.