An average can change substantially when a small dataset contains an extreme value. A trimmed mean responds by excluding a specified portion of the lowest and highest observations before calculating the mean of what remains. The trimming rule is part of the statistic, not a detail to leave out.
NIST’s description of the trimmed mean explains this treatment of the tails and requires the analyst to specify the percentages removed. Terminology can differ over whether a quoted percentage describes the portion retained or removed, so read the method rather than relying on the label alone.
A small example you can check
Consider five invented waiting times, already sorted: 2, 3, 4, 5 and 26 minutes. Their total is 40 minutes, giving an ordinary mean of 8 minutes.
Now apply an explicit rule: remove one observation from each end. The remaining values are 3, 4 and 5. Their total is 12, and their mean is 4 minutes. In this example, the rule removes 20% from each tail, or 40% of the original observations in total.
The median also happens to be 4 here. That coincidence does not make a trimmed mean the same calculation as a median. One averages retained observations; the other identifies the middle of the ordered data under the relevant convention.
Removal from the calculation is not removal from reality
The person who waited 26 minutes still had that experience. A trimmed mean can describe the central part of the data while saying little about the longest waits. If the question concerns service failures, the tail may be exactly where the important story lies.
Imagine two reports about this invented queue. One says “the average wait was eight minutes.” Another says “after removing the shortest and longest waits, the mean was four minutes.” Both can be arithmetically correct. They answer different questions, and neither describes every person’s experience.
Read three lines of method
Look for the sample size, the rule applied to each tail and whether observations carry different weights. Small samples also require a convention when the chosen percentage does not correspond to a whole number of observations.
Finally, ask why trimming suits the question being asked. A clearly stated rule can make a statistic informative and reproducible. An unexplained adjustment can make an ordinary-looking average much harder to interpret.

