We aim to create gradual readability (or target age) measures from infants to elder children. For Japanese texts, several readability measures have been proposed, none of which is applied to texts for infants. Therefore, in this paper, we employ 123 pic- ture books which are clearly decided fine-grained target ages as criterial corpus. Then we investigate the applicability of two previous works to these picture books. Both works show modelate performance. Then we propose a method using new learner and features, and we achieved higher performance to guess the target age.