In this paper we introduced the notion of decision premise

and formed decision implications with decision premises as premises and closures w.r.t.decision subcontext as consequences.It was proven that such decision implications constitute the so-called decision canonical basis

i.e.

it is complete

non-redundant and of minimal cardinality among all complete sets of decision implications.We also described an algorithm to generate decision implication canonical basis and analyzed time complexity of this algorithm.Experiments showed that decision canonical basis can greatly reduce redundant decision implications and is more efficient than other decision implication bases.