Fifty male in-patients receiving treatment for alcohol dependence were studied on admission using the repertory grid and the grids analysed using the principal component technique (INGRIDA). Three self-elements were examined: self-when-sober, self-when-drunk, and ideal self. Sober-ideal self distance was below average in 90 per cent and drunk-ideal above average in 83 per cent, suggesting unrealistically favourable perceptions of sobriety and unrealistically unfavourable perceptions of drunkenness. The percentage of total variation accounted for by the first two components (VAR1 + VAR2) was generally high, suggesting tightness of construing. The first grid did not predict treatment completion. Twenty-eight subjects completed treatment and were given a second grid test on discharge. Consistent changes in grid measures with treatment were not apparent. Forty-nine subjects were followed up six months after discharge. For the 28 who had completed treatment, tightening of construing during treatment (i.e. a higher value of VAR1 + VAR2 in the second grid than in the first) was associated with drinking at follow-up. Measures derived from the first grid alone or from the second grid alone were not associated with status at follow-up.