Chris Snowdon blogs about the "j-curve":
...the optimal level of drinking is quite low, at around one drink per day (or 5-10 units per week), but that you would have to drink a lot more than before your mortality risk rises to that of a teetotaller. If complete abstinence is 'safe', then so is drinking 30, 40 and perhaps 50 units a week. Only beyond this point does one's risk rise above that of the teetotaller.
Go read the rest - very interesting stuff that you won't hear from the Church of Public Health.