This phenomenon is known as metathesis.
Two major hypotheses for the metathesis in this particular case:
Steven Pinker has proposed a phonotactic explanation for the conversion of nuclear to nucular: the unusual and disfavored sequence [kli.ər] is gradually transformed to a more acceptable configuration via metathesis.
However, Arnold Zwicky notes that [kli.ər] presents no difficulty for English speakers in words such as pricklier. He also regards the proposition of metathesis as unnecessary. Zwicky suggests a morphological origin, combining the slang nuke with the common sequence -cular (molecular, particular, etc.). Supporting Zwicky's hypothesis, Geoffrey Nunberg quotes a government weapons specialist: "Oh, I only say 'nucular' when I'm talking about nukes." Nunberg argues that this pronunciation by weapons specialists and by politicians such as Bush – who are aware of the more accepted pronunciation – may be a "deliberate choice". He suggests that the reasons for this choice are to "assert authority" or to sound folksy.
I personally find the Zwicky hypothesis more compelling: the pronunciation is formed by analogy with words like "molecular" and "particular". But both are possible, and in fact, it doesn't necessarily have to be the same method of production for all speakers — there could be interspeaker variation.
(Note that Zwicky's explanation isn't actually "metathesis" per se at all; instead, being nuke + ular, it is formed "as it should be".)