Excel is cartesian, which is hardly 'graphical'. There are virtually no superfluous graphical detail (e.g. placement of shapes, arrows, thickness etc.) -- the only part of the flexgrid that has semantic meaning is the row and column reference.
Giving meaning to coordinates in a 2D chart makes it a graphical language. It certainly is different from conventional languages, which use a tree metaphor (nested expressions) and a linear syntax (everything is considered a long stream of text to be parsed).
Using rows and columns instead of code blocks helps a great deal in reducing the cognitive load of having to address every part of the language through an ID, either for variables of functions. Not to mention intermixing code and data values, which is not considered part of conventional languages (though is common in interactive debugging sessions, which are typically considered "visual" debugging).
