If the data points are the result of experimental data or sensor outputs, it may not be possible to find the exact curve fit, because of measurement/sensor error. However, if you have some knowledge of the mathematical function that the data follows it is possible to find an accurate representation. For example, if the physical process that is represented by the data is known to have an exponential relationship, then an exponential curve fit will give the best results, while a polynomial fit may not be acceptable. Of course, it is always possible to get a polynomial that will pass exactly through all of the n data points by using an (n+1)th order polynomial. However, the curve fit might not necessarily be representative of the underlying function between the measured data points.