Symbolic regression (SR) identifies mathematical equations behind data, which plays a crucial role in scientific discovery. Most SR methods involve coefficient estimation, a process of adjusting the coefficient values in the estimated equation to fit the data. However, existing coefficient estimation methods are prone to fail when estimating exponential and non-exponential coefficients simultaneously. To address this challenge, we propose an algorithm that separately estimates exponential and non-exponential coefficients. Our method finds the ground truth coefficient values in a larger number of problems than existing methods, and the performance can be further improved given an order-level initial estimate of the coefficients.