Introduction: Our purpose was to examine whether the prenatal prediction of a critical birth weight discordance (BWD) in twin pregnancies can be improved by using either different formulae for prenatal fetal weight estimation or single biometric measurements or by assessing the intertwin discordance in the second trimester rather than shortly before birth. Material and Methods: We conducted a retrospective study involving 196 twin pregnancies with an ultrasound examination between 18 and 25 weeks of gestation and one within 14 days prior to delivery. The accuracy was assessed by comparing the prenatal intertwin discordance (PID) with the BWD. PID was estimated by 5 common fetal weight estimation formulae and by single biometric measurements prior to delivery and in the second trimester. Results: The fetal weight estimation accuracy was similar in mono- and dichorionic pregnancies and the smaller and the larger twin. PID was most accurate with the fetal weight estimation formulae prior to delivery. The second-trimester measurements resulted in an underestimation of the BWD. Detection and false-positive rates for a BWD ≧20% were about 65 and 15%. Discussion: About two thirds of the twin pregnancies with a relevant BWD can be detected prior to delivery. An optimal detection rate requires fetal weight estimation close to delivery.