Accuracy class is a standardised way of declaring how far wrong an instrument may be. On an instrument transformer, class 0.2 means the ratio error does not exceed 0.2 % within the range of load and burden it is specified for; on a power quality analyzer the class is defined in IEC 61000-4-30, which distinguishes class A, fit for contractual arbitration, from class S, intended for surveys.
It is worth reading the small print. The class is guaranteed under reference conditions — temperature, frequency, waveform, load — and outside them the error grows. Moreover the error is usually expressed as a percentage of reading plus a number of digits or a percentage of full scale: near the bottom of the range, that second term dominates.
For field work the practical consequence is to choose the right measuring range and not to compare instruments by the class figure alone without looking at the conditions it is given under. An excellent class used at 5 % of full scale can be worse than a modest class used in its optimum band.