Light fastness is one of the most important fastness that should be examined for the dyed textiles which being exposed to light at their end-uses. In this regard,visual assessment of the light fastness is evaluated base on ISO and AATCC standard methodes, and so far a comprehensive and accepted method in industry has not been provided for the instrumental evaluation of light fastness. One of the most important problems in the visual assessment of the light fastness is comparison of the samples with the Blue Scale standard instead of the Gray Scale and the claiming about abnormal color changes due to the operation of the Blue Scale in the exposure device. Therefore, it is possible that observer's experience in the evaluating of light fastness is more important than other color fastness. But in any case, the unavailability and high cost of using experienced observers in the industry is always a problem. Therefore, in the first part of the study, the hypothesis of replacing the mean of repeated visual assessment of the light fastness of inexperienced instead of visual assessment of experienced observers was examined. For this purpose, color samples with different light fastness garde were prepared and method 2 of the ISO 105-B02 standard was used to perform the light fastness test. Visual assessment was performed by 20 inexperienced iranian observers in three repetitions and 5 iranian experienced observers with two repetition. Based on the results obtained from STERESS, PF/3 and root mean square errors, the inexperienced observer's accuracy was improved with the repetition of the assessments possibly due to a better understanding of the experimental protocols and improving their experience. Comparison between the mean of three trials of the inexperienced observers and the results of visual assessment of the inexperienced observers showed that in addition to the high correlation coefficient between assessments, the root of the mean error squares is less than 0.5 and it is an acceptable limit in the industry. Thus, all the results from inexperienced and experienced observers were used to prepare a valid dataset for training a NN and regression. Comparison of neural network and regression results showed that neural network predicted the light fastness of the samples more accurately than the proposed regression methods. The absolute average error rate for the neural network was closer to zero (0.2), and the root mean square errors was acceptable in the industry (0.5), and a complete agreement was obtained between the results.In the final part of the study, an instrumental evaluation was performed to identify the endpoints of the various stages of the light fastness test. In this part of the research, the Blue Scale criteria from 1 to 7 with color difference equal to 2-3 to 5 Gray Scale were used to prepare the samples. Each sample was evaluated by 20 inexperienced Iranian observers, 7 Iranian experienced observers, 7 German experienced observers, and 15 Italian experienced observers. The results showed that one of the proposed methods for correcting GRC formula coefficients resulted in the lowest amount of STRESS, PF/3 and root mean square error, and also the lowest P%. The error obtained from this method for experimental samples was based on the root mean square error (0.23), which was less than the error between the observers and also the error between the different groups of the observers.