I need to test the mean of two very big point shapes. One point shape has around 34.000.000 points and the other around 1.000.000. I have to do a random sample in ArcGIS 10.3.1. Unfortunately, I haven’t found a tool to do this random sample. It should be really randomly and not for example every tenth point. I got these points by running the tool raster to point, so if there is no solution for the point shapes, perhaps there is a straightforward way to randomly sampling a raster?

3 Answers 3


You could use the Create Random Points tool.

Set the Constraining Feature Class to be your points feature class.

The tool will then generate the specified number of points on top of random points from your feature class into a new feature class.

You can then use the newly created points to do a spatial selection of your original points feature class.

  • Let us know how you get on in case someone has the same problem in the future.
    – Dan_h_b
    Commented Apr 13, 2017 at 10:08
  • 1
    Never knew about the constraining feature class...nice!
    – Hornbydd
    Commented Apr 13, 2017 at 13:46

well, if you know how to use R, you can read in the point file, create a random sample of numbers between 1 and 34 million (base::sample), and extract the points by number and write. Eg.

Or in ArcGis; generate your random numbers here, save as a csv. Load the point file and csv into ArcMap, join point data to the csv. If your point file doesn't have xy data then add this first.

  • Thank you so much for your answers. Over the last hours I learned a little bit R, and could do the t-test without taking a random sample. But anyway, it doesn’t make that much sense to test such big data with a t-test… Unfortunately I am not allowed to download the sampling tool at the moment, so I cannot give feedback. Best Nora
    – nora
    Commented Apr 13, 2017 at 18:26

I happend to use the Sampling Design Tool available from here: http://www.arcgis.com/home/item.html?id=ecbe1fc44f35465f9dea42ef9b63e785.

I used it under ArcGIS 10.1. It allows either to draw a random sample (also across different strata), or to subsample from a larger universe of points. It deals with vector data (points). I haven't tested it with such a large sample.

You may want to give it a go. Let us know if that solves you problem.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.