If you plan on comparing NDVI images taken from different dates/times/sensors, then it is a good idea to convert the radiance image to reflectance by performing an atmospheric correction. FLAASH or Quac would probably be the best tools unless you have field data to perform an empirical line calibration. You can also get relative reflectance using tools such as IARR Reflectance, Log residuals, or Flat Field.

Thanks a lot. I will not make compare among dates, times or sensors, just among sites within the same region and within the same image. I don't have field data, so I will use QUAC or FLAASH