There isn't a single best method. It depends on the nature of the data (both in the starless image and the original (star) data). The more the starless image diverges from the original data- the more difficult it becomes of course.
A small menu includes:
1. Writing the pixels from one image to the other using Pixel Math in combination with a mask. This is fraught with issues unless great care is taken to make the two images work well together.
2. Blend the images with:
a. Max (Pixel Math)
b. Lighten (Blend Script or Pixel Math)
c. Opacity (Pixel Math)
d. Screen (Blend Script or Pixel Math)
e. Some conditional statement (iif(something, then something, else $T)... Pixel math obviously
2 d. Is something that I specialize in with the right kind of data. It is a technique not widely implemented... but it is powerful!
-adam