Multiwavelets are briefly reviewed and preprocessing and postprocessing
for such wavelets are introduced. Least squares curve fitting of irregularly
sampled data is achieved by means of unshifted 
and shifted multiscaling functions.
This preprocessing procedure combined with multiwavelet neural networks
for data-adaptive curve fitting is shown to perform well in the case of
high resolution. In the case of low resolution it is more accurate 
than numerical integration and cheaper than matrix inversion.