I've think I've discovered a problem with the Landsat TM pre-processing functionality built into ENVI.  As accessed through the ENVI menu system, the tmcal_doit function does not correctly adjust for the gain/bias as used in the L1T processing for Landsat 5 or 7.  It appears as though it does not know that the minimum QCal DN value is 1, not 0, and miscalculates both the gain and bias when converting to at-sensor-radiance.  This causes errors in both calculations of at-sensor-radiance and top-of-atmosphere reflectance. For that reason, I've created an IDL script to parse the MTL metadata file and for each of the visual, nir- and mid-IR bands do a radiance conversion with gains/biases calculated from the metadata file and then stacks the resulting calibrated data.  It can process L1T data from both Landsat 5 and 7 (and I assume 4 but haven't tested it), and generates correct at-sensor-radiance ranges for non-zero QCal DNs. Is this problem on ITT's radar and will it be fixed in future versions or would there be interest in me cleaning up my processing script to release to the community?