By noting that a Rasch or two parameter logistic (2PL) item belongs to the exponential family of random variables and that the probability density function (pdf) of the correct response (X=1) and the incorrect response (X=0) are symmetric with respect to the vertical line at the item location, it is shown that the conjugate prior for ability is proportional to [I(theta)]superscript alpha where I(theta) is the item information and alpha is a positive constant. When the above prior is applied to a three parameter logistic (3PL) item, the requirement that item selection rules are bound to the traditional formula for correction for random guessing implies that the constant alpha must be 1. Thus, maximum information selection rules for 3PL items are the only rules that are consistent with a Bayesian analysis based on the family of conjugate priors and with the use of the correction-for-guessing formula. (Contains 16 references.) (Author/SLD)