In his survey article, Gordon conjectured that
all Seifert fibered surgeries arise
from Dean's primitive/Seifert-fibered construction.
We construct two infinite families of knots each of which admits
a Seifert fibered surgery
with none of these surgeries coming
from Dean's primitive/Seifert-fibered construction.
This disproves the conjecture.
The starting point is the $(-3,3,5)$ pretzel knot which
belongs to both of the infinite families.