To create a detailed building energy model, the modeler must choose dozens or hundreds of parameter values to describe the thermal properties of the building envelope, the efficiency of the heating and cooling system, the temperatures maintained in various parts of the building at various times of day, and so on. Usually some of these values are not precisely known. Typically, a modeler makes her best guesses, runs the models, compares the model output to data --- which typically includes heating, cooling, total power, and room temperatures as a function of time --- and attempts to adjust the model to better fit the data. This hand-tuning can be quite difficult and time-consuming. The speakers have created software tools, collectively known as Software for Tuning Energy Models (STEM), that largely automate model tuning. It has been applied only to Energy Plus but could be used with other building energy modeling programs. STEM adjusts parameter values, within modeler-determined limits, to minimize an error metric that can include almost any available building data; typically, this is a combination of the root-mean-squared errors in cooling and heating energy, lighting, plug loads, and room temperatures. A parallel version of the software, running at NERSC, can optimize 70 parameters for a season-long Energy Plus run of a large building (the UC Merced Classroom and Office Building) with complicated solar shading in about 24 hours of clock time; on a single-processor machine, such a large optimization takes about a week. Results are much faster when working with just 5 or 10 parameters. In this talk, we will discuss the benefits and limitations of the software, with examples from the UC Merced Classroom and Office Building. We will discuss approaches to dealing with unknowns such as schedules and occupancy. We will also touch on issues such as overfitting and model misspecification errors.