Functional magnetic resonance imaging of the human brain in action presents
large statistical and computational challenges. Here we describe those cha
llenges and provide references to a number of other papers where detailed m
ethods developed to meet them are reported. The first seven sections of thi
s paper were written in 1995 when our work was in its infancy. The last fou
r sections were written more recently, to update the earlier sections and t
o show the directions we have gone and the directions we intend to go.