The term pure function used in Mathematica is not being used in the same sense as the cited Wikipedia article. In Mathematica it refers to an anonymous function. In the Wikipedia article it is a term extracted by analogy from the increasingly popular term "purely functional" which refers (mainly) to deterministic programming free of side-effects.
The Mathematica usage largely predates the contemporary popular usage.
Pure Function in Mathematica
Section 4.1.4 of the Mathematica Book (1988), states:
There are various ways to specify the functions you give as arguments. If they are built-in functions, or are functions that you have explicitly defined, then you can refer to them simply by giving their names. In many cases, however, it is more convenient to be able to build up "pure functions" to which you do not have to give specific names.
This usage aligns with what are often called "anonymous functions" today. Note the use of quotation marks around "pure functions" suggesting that the term is being coined here. Later, it is said that:
This kind of Mathematica expression is directly analogous to the λ expressions that are used in LISP or in formal logic.
Again, this emphasizes their usage as anonymous functions. And then:
One goal of functional programming is completely to eliminate named variables.
So, in Mathematica, it appears that the "purity" of a function means that is it unencumbered by names. (I note in passing that eliminate named variables suggests the point-free programming style to me. Slots in pure functions do not conform to that style, but V10 is adding features that make point-free style a little easier.)
Pure Function in Wikipedia
The cited Wikipedia article defining the term "pure function" is tagged as controversial for lack of supporting citations. Neither of the two references that it cites define, or even mention, the term "pure function".
They do, however, speak of "pure functional" or "purely functional" programming languages. I suspect that "pure function" in that article is simply a coined reduction of the far more widespread term "purely functional".
Purely Functional in Popular Usage
My own impression is that the "pure" qualifier with respect to functional programming was relatively obscure until the Haskell programming language started to get some mindshare. Haskell emerged from the pure functional programming community in 1990, two years after The Mathematica Book, and even then did not really gain popularity until after the release of Haskell98.
I find it interesting to note that the following papers make no use of the term "pure" in connection with functional programming:
But we start to see usage of the term by the time of these:
If the Google books Ngram Viewer is to be believed, then the Mathematica usage of "pure" appears at the leading edge of an upsurge in wider usage:

Summary
I interpret this (admittedly anecdotal) evidence to support the position that the Mathematica use of "pure function" is not a misapplication of the general usage. Rather, it is an independently coined term for "anonymous function" that predates the contemporary common understanding.
=
or:=
associates a transformation rule with the symbol on the LHS, e.g.OwnValues
,DownValues
,SubValues
, etc. So, in that sense, it is pure. $\endgroup$ – rcollyer Oct 31 '14 at 15:26pure-function
here on MM.SE says "Questions about implementing or applying functions which: (1) always yield the same result for the same input, (2) are independent of outside information, and (3) produce no side effects." which is again not the Mathematica definition $\endgroup$ – masterxilo Aug 3 '16 at 15:54