The transient problem in self-tuning control is further analysed in a non-s
tochastic framework based on model unfalsification and solutions are propos
ed. It is not assumed that the noise bounds or norm bounds on the unmodelle
d dynamics were known a priori. Depending on the transient requirements on
output bound, tracking performance and length of tuning period, three cases
are analysed. Simulations show the practical potential of the tuning metho
ds presented.