Unconstrained Handwritten Numeral Recognition Has Been A Recent Research Area From Last Few Decades. Handwritten Numeral Recognition Approach Is Used In Many Fields Like Bank Checks, Car Plates, ZIP Code Recognition, Mail Sorting, Reading Of Commercial Forms Etc. This Paper Presents A Technique To Recognize Handwritten Numerals, Taken From Different Pupils Of Different Ages Including Male, Female, Right And Left Handed Persons. 340 Numerals Were Collected From 34 People For Sample Creation. Conjugate Gradient Descent Back-propagation Algorithm (CGD-BP) Is Used For Training Purpose. CGD-BP Differs From Primary Back-propagation Algorithm In The Sense That Conjugate Algorithms Perform Line Search Along Different Directions Which Produce Faster Convergence Than Primary Back Propagation. Percentage Recognition Accuracy (PRA) And Mean Square Error (MSE) Have Been Taken To Estimate The Efficiency Of Neural Network To Recognize The Numerals.