Garch fitting when some of the data might be inaccurate

Can anyone guide me to any newbie-ish literature on how to fit garch
parameters (for example in Garch 1,1) to data where a small percentage
of the data (perhaps 0.1%) might be missing or inaccurate.

As I understand it, maximum likelihood techniques are not at all
robust with respect to any erroneous or missing data.

I currently use matlab's garch toolbox to determine the values of the
garch parameters.

Many thanks for your help