In the original modern Beauty Contest game a player has to choose a number between zero and 100 with the objective to be closest to 2/3 times the average with a fixed prize payment. Zero is the unique rationalizable equilibrium. Behavior, however, is very heterogeneous since many subjects wrongly anchor their choices (described by a level k model). This game provides an important element in macroeconomic models which are frequently used for making policy decisions, e.g. the new Keynesian model, neo classical models, global game related models etc. Thus, the policy maker has to be concerned about whether actual behavior is in equilibrium or whether there is some convergence to equilibrium. This is because policies that give good results in equilibrium, they may not work so well, for example, if people act like level-k players. We provide a new model with two substantial changes of the game, one which we believe reflects better what happens in reality: one is that each player receives an idiosyncratic signal drawn from a normal distribution about the key variable to predict; the other is that the action is not restricted to a bounded interval. The aggregate theoretical prediction does not change with this new game, but we show that laboratory experimental behavior is in closer agreement with the theory, even in less sophisticated subject pools, whose behavior in this game is similar to the one of sophisticated groups, such as economic professors, in the traditional game. We also provide comparisons of the total surplus achieved in this game with respect to what happens in the traditional one.