The effects of data pretreatment and mean levels upon nonlinear model
structures is investigated. Techniques commonly used to prefilter data
in linear system identification are shown to alter the model structur
e in the nonlinear case. The effects of mean levels are considered in
detail and a new unravelling algorithm is derived to recover the under
lying system model when the offsets are external to the system. A new
mapping from the time domain to the frequency domain is also introduce
d for the case where offsets can be considered as an implicit part of
the system.