Handling spatial reference

If all of your data is in the same coordinate system, it is best to perform analysis in that coordinate system. However, the output spatial reference system of any Spatial Analyst object can be specified using the analysis environment by setting the OutSpatialReference. If the output spatial reference system is not specified, a Spatial Analyst operation uses the spatial reference of the first raster input that has a spatial reference other than Unknown.
 
If no raster inputs have spatial reference information, the spatial reference is taken from the first feature input with a defined spatial reference. If no input dataset has a defined spatial reference, the output has an unknown coordinate system. Any other input with a spatial reference different from the output is projected on-the-fly into the output coordinate system before performing the operation.
 
For vector data, this projection on-the-fly is very accurate; however, raster data is projected using an algorithm that produces faster results than a true projection at the cost of high accuracy. For raster datasets at low latitudes (less than 50 north or south) and smaller than a few degrees on each side, this projection is fairly accurate, but this projection is inappropriate for larger or polar raster datasets. In these cases, project your input datasets beforehand to avoid projection on-the-fly for highest accuracy.

For more information on geodata transformations and accurate projections, see DataSourcesRaster.