This paper provides new evidence on bias in preference revelation and valuation of public goods. It does so through the systematic definition of free-rider, mixed-good, information and social choice problems and by a survey application to the case of valuing social benefit from the arts. These sources of bias are each found to be significant. It is also found that free-rider behaviour is random with respect to observable socio-demographic characteristics. Allowing for bias, aggregate public good benefit necessary for public support of arts is still established. The study therefore demonstrates not only the significance of free-rider and other biases in eliciting public good demand, but also practicable techniques for allowing for these effects via a survey method.