The paper proposes a new approach for function approximation by 'estimating' the second derivative, using Taylor series, where the samples of the function and its first derivatives at the sample points are known. Without such 'estimation', these initial data, can approximate the function traditionally using the first order Taylor approximation, but with more error. If we desire to improve the approximation via second order Taylor series, then we can estimate the 'pseudo' second derivatives of the function in three different ways. All these three ways are investigated in this paper. The pseudo second derivatives help in computing many more sample points of the function within each sampling interval. Thus, the approach acts like a mathematical 'magnifying glass'. Two examples are treated to compare the efficiencies of the methods. Also, a qualitative study for upper bound of error of the approximations is studied in detail. © 2016 Elsevier Inc. All rights reserved.