This paper investigates information acquistion in decision problems. We introduce a new notion of "better information", Accuracy-order (A-order), defined on continuous families of signals. Accuracy formalizes the idea that "a signal that is more correlated with the unknown random variable is better". This concept is indigenous to an economically interesting subset of all decision problems, those where signals are affiliated and the payoff function satisfies the single-crossing property. On this subset, this notion is found to be "tight", in the sense that A-order is an if-and-only-if condition for better information. Thus, a Blackwell-type result is obtained. On the subset, it is shown that Blackwell's Sufficiency is a special case of Accuracy. Finally, a comparative statics result is obtained, about which decision problem will induce more information acquistion.