Thyroid nodules are solid or fluid-filled lumps in the thyroid that affect 70% of the Mexican population. These nodules can be malignant, so identifying techniques for their classification is very important. As a result, multiple studies have been conducted to analyze ultrasound images. Current research shows that the best results for thyroid nodule classification involve feature extraction from images, especially using wavelet decomposition. However, access to large datasets has only recently become possible, so few studies have explored wavelet-based features for classification, resulting in limited information on this technique. Therefore, this work focuses on finding a mother wavelet on first-level of decomposition that provides the features that better differentiate, with a statistical significance difference, between benign and malignant thyroid nodules. For this research, a database with 17,412 ultrasound images with their corresponding segmentation masks was used. First-level wavelet decomposition was applied using five mother wavelets (Daubechies8, Haar, Symlet2, Beylkin, and Fejér-Korovkin4). Eight first-order features were extracted from the wavelet coefficients, which resulted in feature vectors for both benign and malignant thyroid nodules. Mann-Whitney U test and Cliff’s Delta were used to detect and quantify, respectively, the differences between the feature vectors of benign and malignant nodules. The statistical analysis indicates that, at this level of decomposition, Symlet2 and Fejér-Korovkin4 mother wavelets offer the best features for thyroid nodule classification.