Ventriculoperitoneal (VP) shunt insertion is a common neurosurgical procedure, essentially unchanged in recent years, with high revision rates. We aimed to identify potentially modifiable associations with shunt failure. One hundred and forty patients who underwent insertion of a VP shunt from 2005-2009 were followed for 5-9years. Age at shunt insertion ranged from 0 to 91years (median 44, 26% <18years). The main causes of hydrocephalus were congenital (26%), tumour-related (25%), post-haemorrhagic (24%) or normal pressure hydrocephalus (19%). Fifty-eight (42%) patients required ⩾1 shunt revision. Of these, 50 (88%) were for proximal catheter blockage. The median time to first revision was 108days. Early post-operative CT scans were available in 105 patients. Using a formal grading system, catheter placement was considered excellent in 49 (47%) but poor (extraventricular) in 13 (12%). On univariate analysis, younger age, poor ventricular catheter placement and use of a non-programmable valve were associated with shunt failure. On logistic regression modelling, the independent associations with VP shunt failure were poor catheter placement (odds ratio [OR] 4.9, 95% confidence interval [CI] 1.3-18.9, p=0.02) and use of a non-programmable valve (OR 0.4, 95% CI 0.2-1.0, p=0.04). In conclusion, poor catheter placement (revision rate 77%) was found to be the strongest predictor of shunt failure, with no difference in revisions between excellent (43%) and moderate (43%) catheter placement. Avoiding poor placement in those with mild or moderate ventriculomegaly may best reduce VP shunt failures. There may also be an influence of valve choice on VP shunt survival.