Seven cases are described in which neurological damage followed spinal or combined spinalβepidural anaesthesia using an atraumatic spinal needle. All patients were women, six obstetric and one surgical. All experienced pain during insertion of the needle, which was usually believed to be introduced at the L~2β3~ interspace. In all cases, there was free flow of cerebrospinal fluid before spinal injection. There was one patchy block but, in the rest, anaesthesia was successful. Unilateral sensory loss at the levels of L~4~βS~1~ (and sometimes pain) persisted in all patients; there was foot drop in six and urinary symptoms in three. Magnetic resonance imaging showed a spinal cord of normal length with a syrinx in the conus (nβ=β6) on the same side as both the persisting clinical deficit and the symptoms that had occurred at insertion of the needle. The tip of the conus usually lies at L~1β2~, although it may extend further. Tuffier's line is an unreliable method of identifying the lumbar interspaces, and anaesthetists commonly select a space that is one or more segments higher than they assume. Because of these sources of error, anaesthetists need to relearn the rule that a spinal needle should not be inserted above L~3~.