A cylinder is fine, it is lines that you should be wary of. Technically, there is no such thing as a line - it is either an axis of revolution (cone or cylinder axis for example) or a linear cross section of a planar surface. Obliviously, as CMM programmers, we use lines all the time and have become accustomed to having to create constructed features of various types in order to work around software limitations and apply what we think is the correct design intent. The Geometric Tolerance command requires a different way of thinking and working which greatly simplifies things. Constructed features are generally not needed and can actually hinder the way the Geometric Tolerance command is designed to work because they obscure access to the hit data. Rather than relying on the user to interpret design intent and create constructions and alignments, the Geometric Tolerance command simply requires you to supply hit data and build the feature control frame - it handles the rest. The best way to ensure you are supplying good hit data is to measure simple features (auto-features, measured features and scans).
To come back to your question about using a cylinder and why that differs from a line, the distinction I'm making is that the line only represents the axis, a cylinder represents the surface as well. The Geometric Tolerance command simulates hard-gauging and requires surface data. The accuracy of the result and how closely that result will correlate with an actual hard gauge check is dependent on how much of the surface you have probed and supplied data for.
