Prediction of the first visibility of the lunar crescent is a difficult problem involving astronomy, meteorology, and physiology. Historically, this problem has been attacked by an empirical approach where some set of observations is used to deduce a criterion for visibility. In this paper, I present a list of 201 observations and their observing circumstances for use in deriving and testing prediction algorithms. I find that criteria involving the moonset lagtime and the Moon’s age are quite bad in their predictive ability. Criteria involving the relative altitude and azimuth of the Moon at sunset are better, yet still can yield incorrect predictions within a zone of uncertainty with a width of over 105 degrees in longitude. The new theoretical model of Schaefer ( 1 988) is found to have a zone of uncertainty with an average total width of 47 degrees in longitude.