GAUSS2DFIT with the /NEGATIVE option is supposed to fit a gaussian "valley" instead of "peak."  
 i came across intermittent cases where the function seemed to give a good fit in terms of parameter  
 values, except that the output parameter A[1] was positive instead of negative. also, the return  
 value had a positive gaussian peak instead of a negative gaussian valley. 
  
 i have not thoroughly tested this, but adding this line of code seems to fix it: 
  
     a[1] *= keyword_set(neg) ? -1 : 1 
  
 this should be placed on line 263, just before the comment "If there's a tilt". 
  
  
  
  
  
  
 this is for IDL Version 8.5.1, Mac OS X (darwin x86_64 m64). 
 please disregard if you are using a later version where this has already been fixed.