The effects of relying on outmoded IQ-test norms and the use of arbitrary classifications of developmental delay on estimates of cognitive impairment of very preterm infants (VPI) was evaluated in a prospective population study. Cognitive assessments included the Griffiths test at 5 and 20 months and the Columbia Mental Maturity Scales (CMM) and a vocabulary test (Aktiver Wortschatz Test, AWST) at 56 months of age. Rates of cognitive impairment of 321 very preterm infants (VPI; < 32 weeks gestation or < 1500 g birth weight) were determined according to the published test norms, to scores of a full-term control group (FC n = 321), and to scores from a representative sample of children (NC n = 431) of the same birth cohort. IQ-scores were higher in the FC and NC children than in the original standardisation sample (SS). Using the concurrent test norms (FC, NC) up to 2.4 times more VPI were identified as seriously impaired (<-2 SD) than if the published (outdated) norms were used. Serious developmental delay was underestimated when arbitrary (e.g. DQ < 70) rather than across age comparable definitions (DQ <-2 SD) were used. VPI study drop-outs had mothers with lower educational qualifications and poorer cognitive developmental scores at 5 or 20 months of age. In conclusion, a lack of appropriate control groups and use of arbitrary criteria for judging serious delay leads to large underestimations of cognitive impairment in VPI. Findings from previous uncontrolled studies of VPI need re-interpretation.