In American English, for a normal pronunciation with only the middle vowel stressed, it's ba-nan-a. However, for a slowed up, syllable by syllable pronunciation, where all three vowels are stressed, it's ba-na-na. The evidence for this comes from the flapping phonological process, which changes intervocalic syllable final alveolar stops in most American dialects into flaps. The consonant n is an alveolar stop (in the sense that there is a momentary complete obstruction in the mouth to the air stream).
In "banana", the first n is not flapped and the second one is. That tells us that the first n is at the beginning of a syllable, and the second n is at the end of a syllable. Thus, the syllabification must be ba-nan-a. The reason for this syllable division is that stressed vowels in English attract consonants into their own syllables, while unstressed vowels tend to reject preceding consonants from their own syllables. So the second "a" which is stressed wants the preceding n and the following n both to go in its own syllable.
However, if you are sounding the word out by syllables, for clarity, and you stress all three vowels, then the last two vowels pull the preceding n's into their own syllables, and the last vowel, now stressed, does not reject the preceding n from its own syllable.