Clinical trials show that protein supplement increases infant size in malnourished populations; however, epidemiological studies in high-income countries have reported mixed results. Although these findings suggest a non-linear relationship between maternal macronutrient intake and fetal growth, this relationship has not been closely examined. We assessed the association between maternal protein intake and fetal growth among 91 637 Japanese women with singletons in a nation-wide cohort study using validated FFQ. The respondents answered the FFQ twice, once during early pregnancy (FFQ1; 16·3 (sd 6·0) weeks), and second during mid-pregnancy (FFQ2, 28·1 (sd 4·1) weeks). Daily energy intake and percentage energy from protein, fats and carbohydrates were 7477 (sd 2577) kJ and 13·5 (sd 2·0), 29·5 (sd 6·5) and 55·3 (sd 7·8) %, respectively, for FFQ1, and 7184 (sd 2506) kJ and 13·6 (sd 2·1), 29·8 (sd 6·6) and 55·3 (sd 7·9) %, respectively, for FFQ2. The average birth weight was 3028 (sd 406) g, and 6350 infants (6·9 %) were small for gestational age (SGA). In both phases of the survey, birth weight was highest and the risk of SGA was lowest when the percentage energy from protein was 12 %, regardless of whether isoenergetic replacement was with fat or carbohydrates. Furthermore, when protein density in the maternal diet was held constant, birth weight was highest when 25 % of energy intake came from fat and 61 % came from carbohydrates during early pregnancy. We found maternal protein intake to have an inverse U-curve relationship with fetal growth. Our results strongly suggest that the effect of protein on birth weight is non-linear, and that a balanced diet fulfilling the minimum requirement for all macronutrients was ideal for avoiding fetal growth restriction.