Piecewise versus fitting to a distribution

#1
Question: When defining a set of data...is it "better" or more valid to use actual percentages within a set range or to fit the data to a distribution using a statistical package? What are the tradeoffs?