Purpose This study aims to evaluate the accuracy of lens prediction formulae on a paediatric population. Methods A retrospective case-note review was undertaken of patients under 8 years old who underwent cataract surgery with primary lens implantation in a regional referral centre for paediatric ophthalmology, excluding those whose procedure was secondary to trauma. Biometric and refractive data were analysed for 43 eyes, including prediction errors (PE). Statistical measures used included mean absolute error (MAE), median absolute error (MedAE), Student's t-test and Lin's correlation coefficient. Results The mean PE using the SRK-II formula was +0.96 D (range − 2.47D to +2.41 D, SD 1.33 D, MAE 1.38 D, MedAE 1.55, n = 15). The mean PE was smaller using SRK/ T (−0.18 D, range − 3.25 D to +3.95 D, SD 1.70 D, MAE 1.30 D, MedAE 1.24, n = 27). We performed an analysis of the biometry data using four different formula (Hoffer Q, Holladay 1, SRK-II and SRK/T). Hoffer Q showed a smaller MedAE than other formulae but also a myopic bias. Conclusion Our clinical data suggest SRK/T was more accurate in predicting postoperative refraction in this cohort of paediatric patients undergoing cataract surgery. Hoffer Q may have improved accuracy further.