This research asks how young children become bilingual, and what best predicts bilingual language development. All mothers of new babies on Anglesey in North Wales were contacted and asked to complete a questionnaire for the family concerning their past and present use of Welsh and English, and their attitudes towards these languages. Use was. taken as more important than knowledge and respondents (N= 413), were allocated to five language background types on the basis of language use. Ten firstborn children with both parents resident were chosen to represent these groups, and recordings were made of their language development at three monthly intervals from age 16 to 36 months. Nine sessions took place at home, most during free play between mother and child, the last between fathers and children at three. This small sample allowed close scrutiny of the process of language acquisition. Families who replied to the first questionnaire were sent a second three years later. This asked about current parental language use and attitudes, and about the development of their child's Welsh and/or English. More than two thirds of families on Anglesey use Welsh and the large majority of families want their children to learn Welsh at school, English-speaking families giving mainly instrumental reasons and Welsh-speaking families mainly integrative reasons. Development in this large group paralleled that of the small sample. It is suggested that children who are becoming bilingual learn their languages sequentially, and an extension to the Threshold Model is proposed. Men are shown to influence the language spoken at home more than women, but the English language has the greatest effect. Children from Welsh-speaking homes are more likely to become bilingual. Although fathers influence their children's language, by far the greatest predictor of future language use is the mother's language when the child is born.