In an urban wastewater treatment plant (WWTP), early knowledge of the pollutant load levels throughout the plant is key to optimize its processes and achieve better purification levels. Molecular spectrophotometry has begun to gain prominence in this wastewater characterization process, as it is a simple, fast, inexpensive and non-invasive technique. In this research work, different mathematical models based on genetic algorithms have been developed for the estimation of chemical oxygen demand (COD) and total suspended solids (TSS) from the spectral response of the samples, measured in the 380-700 nm range by means of a light-emitting diode (LED) spectrophotometer developed by the researchers. A field campaign was carried out in Mapocho-Trebal WWTP (Chile), where 550 samples were obtained in three different parts of the plant: at the inlet (raw wastewater), at the outlet (secondary treated wastewater) and at the outlet of the primary clarifier. A total of 18 estimation models have been calculated by mean of HeuristicLab software, which have presented a high accuracy, with a Pearson's coefficient between 80 and 90% in most cases. In order to achieve the most accurate models possible to characterize each part of the plant, specific models have also been developed, as well as combined models that are valid for all types of wastewater.