Economical adjunction of square roots to groups

How do we have to extend a group so that in the resulting group all elements of the original group be squares? We give a rather precise answer to this question (the best possible upper bound differs from our estimate by at most a factor of two) and pose several open questions.