This paper compares the use of perceived and measured noise in a hedonic housing model. Although in theory the use of subjective variables is recommended, most empirical applications use measured noise variables. Merging different databases, we obtain a sample of about 2,800 apartments located in Geneva, Switzerland, containing both measured and perceived noise data. We make the measured and perceived noise data comparable and analyse their performance by fitting alternatively three hedonic models. We find that for moderate to high noise levels the measured noise safely approximates its perception.