Detection of refractive error in children is crucial to avoid amblyopia and its impact on quality of life. We here performed a retrospective study in order to develop prediction models for spherical and cylinder refraction in children. The enrolled 1221 eyes of 617 children were divided into three groups: the development group (710 eyes of 359 children), the validation group (385 eyes of 194 children), and the comparison group (126 eyes of 64 children). We determined noncycloplegic and cycloplegic refraction values by autorefractometry. In addition, several noncycloplegic parameters were assessed with the use of ocular biometry. On the basis of the information obtained from the development group, we developed prediction models for cycloplegic spherical and cylinder refraction in children with the use of stepwise multiple regression analysis. The prediction formulas were validated by their application to the validation group. The similarity of noncycloplegic and predicted refraction to cycloplegic refraction in individual eyes was evaluated in the comparison group. Application of the developed prediction models for spherical and cylinder refraction to the validation group revealed that predicted refraction was significantly correlated with measured values for cycloplegic spherical refraction (R = 0.961, P < 0.001) or cylinder refraction (R = 0.894, P < 0.001). Comparison of noncycloplegic, cycloplegic, and predicted refraction in the comparison group revealed that cycloplegic spherical refraction did not differ significantly from predicted refraction but was significantly different from noncycloplegic refraction, whereas cycloplegic cylinder refraction did not differ significantly from predicted or noncycloplegic values. Our prediction models based on ocular biometry provide estimates of refraction in children similar to measured cycloplegic spherical and cylinder refraction values without the application of cycloplegic eyedrops.