The Last Digit test is a notable method to detect election fraud. It is based on an assumption that a manipulator replaces the vote counts of an election result sheet with man-made numbers, but will fail to make the numbers look random. Allegations of election fraud are based on this mechanism, but the strategic behaviour of humans when manipulating election results could make it difficult to detect fraudulent activities. This paper is the first to use a laboratory experiment to evaluate the ability of the Last Digit test to detect human manipulation of election results. Only a small share of participants’ manipulations are detected by the Last Digit test. The small sensitivity is due to the strategic behaviour of participants. Participants in the experiment manipulate as few polling stations as necessary to reach their manipulation aim, manipulate leading digits when the requested intensity of fraud is extremely high, and use alternative strategies that are difficult to detect using the test. The analysis of the experiment further shows that only if participants alter a substantial share of last digits in the experiment the test indicates fraudulent activities.