OBJECTIVES: To identify which type of smoke alarm is most likely to remain working in local authority inner city housing, and to identify an alarm tolerated in households with smokers. DESIGN: Randomised controlled trial. SETTING: Two local authority housing estates in inner London. PARTICIPANTS: 2145 households. INTERVENTION: Installation of one of five types of smoke alarm (ionisation sensor with a zinc battery; ionisation sensor with a zinc battery and pause button; ionisation sensor with a lithium battery and pause button; optical sensor with a lithium battery; or optical sensor with a zinc battery). MAIN OUTCOME MEASURE: Percentage of homes with any working alarm and percentage in which the alarm installed for this study was working after 15 months. RESULTS: 54.4% (1166/2145) of all households and 45.9% (465/1012) of households occupied by smokers had a working smoke alarm. Ionisation sensor, lithium battery, and there being a smoker in the household were independently associated with whether an alarm was working (adjusted odds ratios 2.24 (95% confidence interval 1.75 to 2.87), 2.20 (1.77 to 2.75), and 0.62 (0.52 to 0.74)). The most common reasons for non-function were missing battery (19%), missing alarm (17%), and battery disconnected (4%). CONCLUSIONS: Nearly half of the alarms installed were not working when tested 15 months later. Type of alarm and power source are important determinants of whether a household had a working alarm.