Hi, You might try plotting to the Z-buffer device, read the resulting image with TVRD() and then change back to the regular graphics device to display the resulting plot image. For example: set_plot, 'z' plot, a, b, linestyle=1,psym=3 data = tvrd() set_plot, 'win' tv, data I hope this will help. Regards, Jim (Exelis VIS)

I'm sure what ENVI does, and what I would certainly do if I was going to make a scatter plot of two very large images, is to just take a much smaller random sampling of the pixel locations. Otherwise, your plot will look like something the dog did. s = Size(NIRImage, /Dimensions) indices = Round(RandomU(seed, 5000) * (s[0]*s[1])) cgScatterPlot, NIRImage[indices], REDImage[indices], PSYM=3 Choose the number (e.g., 5000) to give you the best combination of speed and information.

I like David's suggestion, but what I might think to do instead (depending on what is in the images) is bin the large images (one pixel for every 5x5 window for instance) to obtain a subset, vs using an entirely random subset.

This was my first thought, too. But, when I am choosing not to display all the data I have, I think there is less chance to trick myself with a random selection of data, then there is with a systematic selection of data like this. Or, another way to say this, I think you will determine if your method is actually working correctly and giving you the results you expect much faster when you use a random selection. Maybe this is just a personal preference, but I choose it to avoid making systematic selection errors. Another argument is that your "grid" method only works correctly for uniformly sampled grids. You will severely under-represent dense areas of the scatterplot and over-represent areas where you have fewer points. In a random selection, you should sample both areas of the image correctly.