The reported investigation is one part of a project concerning methods for measurement of the longitudinal roughness of roads and the necessary accuracy. In this study the main focus was on the subjective experience of roughness on roads with low international roughness index (IRI) values, that is, fairly good roads. With the available data it was also studied how much a random error added to the IRI values would influence the correlations with the subjective estimates. The investigation was carried out as a magnitude estimation experiment. Twenty-two observers made their estimates while traveling as passengers first in a car and later in a truck. The roughness estimates were made on 45 sections along a 60-km route. Most of the stretches had an IRI roughness between 0.5 and 3.0 mm/m, with a few of up to IRI = 5.5. The reference section had an even higher roughness, IRI = 6.24, which was given the nominal subjective roughness magnitude of 100. The main results of the study were as follows: subjective roughness seems to be a linear function of roughness according to IRI within the studied roughness range; for some road sections with a nontypical spectral composition of the road roughness, it was found that the correlation between IRI and subjective roughness decreased considerably, and the simulations of random errors added to the IRI values showed that within the studied range and with the fairly large number of observations (45), random measurement errors up to at least +/-0.2 IRI unit (mm/m) can be considered insignificant.