Tolerance: A tolerance is the allowed variation in a dimension, such as 10 mm plus or minus 0.1 mm, telling the shop how close to the nominal size a feature must be for the part to pass.
No process makes a part to exactly the size drawn. A tolerance says how much difference is acceptable. It is written as a nominal size with limits: 25.00 ± 0.05 mm means anything from 24.95 to 25.05 mm passes.
When a drawing does not state a tolerance on every dimension, a general tolerance applies, often taken from a standard such as ISO 2768. Specific tighter tolerances go only where the function needs them: a bearing seat, a mating face, a hole for a dowel pin.
Tolerances drive cost because tighter limits need slower cutting, more careful setup, finer machines, measurement and sometimes extra finishing operations. A 3D print without finishing usually holds a coarser tolerance than a machined part.
Example: a shaft diameter of 20 mm with ± 0.1 mm can be turned quickly and checked with a calliper. The same diameter at ± 0.01 mm may need finishing passes or grinding and measurement with a micrometer, and the price rises accordingly.
The usual mistake is putting tight tolerances on every dimension “to be safe”. That makes every feature expensive when only one or two need precision. Another is confusing tolerance with surface roughness: a part can be within size and still have a rough surface, or smooth and out of size.
For form, orientation and position, see GD&T. Read tolerances and cost for how tolerance choices move the price.
Go deeper: How tolerances drive machining cost (and when to loosen them)