Bar charts are the default for comparing amounts across categories, because length is easy to judge.
Each bar stands for one category, and its length from 0 is the value.
That reliance on length is why the axis must start at 0: a bar cut short no longer shows its value.
Sorting the bars by value lets the reader see the ranking without comparing every pair.
Alphabetical order is better only when readers need to look up one category by name.
Grouped bars put several series side by side within each category, such as sales for 2 years in each region.
Stacked bars pile series on top of each other, which shows totals well but makes the middle pieces hard to compare.
Only the bottom piece starts from a shared baseline, so the others float at different heights.
Labelling values directly on the bars can remove the need for gridlines and a value axis.
Too many categories make bars thin and labels unreadable, so the smallest are often grouped as Other.
3D bars distort lengths through perspective and add nothing, so they are best avoided.
Bars are a poor fit for a distribution of values, because one bar per group hides the spread inside it.
In practice horizontal bars are often easier to read, since long labels sit on the left at a normal angle.
Colour can highlight the one bar the message is about while the others stay grey.
The usual mistake is a bar axis that starts above 0, which turns small differences into dramatic ones.