Many experiments and even published results from computer systems research use flawed statistics. In this talk, Yan Li will start by briefly reviewing the basic concept of confidence interval and how to calculate it, then proceed to discuss the test of significance for comparison of benchmark results. Understanding how to compare benchmark results is very important because many changes in a computer system lead to only small changes in performance, and you need to use the proper statistics tool to decide if these small changes are real or just caused by random errors. In most cases, you do not have a whole day for this so the benchmarks have to be as short as possible. As you will see in the talk, properly carrying out these methods is tedious and error prone, so Yan Li will introduce the Pilot tool that the SSRC is developing and show how it can help you to automate these tasks.