It is suggested that the peculiar behaviour of VNM utility function in the Friedman-Savage case could be explained by introducing the cost of information into our calculation. To obtain utility from money one needs much information and it often costs money to get information. More money enables one to get more information, and information in its turn increases the utility of the same money. How this process can give in some portion of the curve while retaining the usual in other has been investigated here. © 1985.