In a two-tailed distribution, the tails are the light-yellow regions of the graph, on the right. Imagine that a container of margarine has been filled with 250g, but within an error + or - 2.5g. The container is to be rejected for sale if the error exceeds 2.5g.
This image appears in the article, but it seems to be not relevant to this context of hypothesis testing. There is no hypothesis, and there is no data set other than the single data point of this one container. So I'm going to delete the image. Loraof (talk) 17:29, 21 February 2017 (UTC)