Projective methods are qualitative tools used to study food consumer behavior. In recent years, there has been an increase in studies that use these tools to analyze consumer behavior, particularly with the word association (WA) technique. However, one of the challenges in using these methods is the stability and validity of the data. This research aimed to obtain preliminary information on the stability and validity of the associations generated by consumers with the WA technique, using the structural approach of social representations. For this, two studies were carried out; for the first study, a face-to-face survey was carried out in which 89 consumers participated, who wrote the first words that came to mind with the stimulus “Aculco” on a ballot paper. For the second study, 122 consumers completed the same task as in the first study; however, the participants were recruited from an online survey. A random sample (n = 50) of both studies was selected to explore the stability and validity of the results. In the three study samples, the words were grouped into categories and analyzed through the structural approach of social representations. The frequency of mention of the identified categories was compared with the chi-square test and the average position (AP) and the Cognitive Salience Index (CSI) were calculated. Prototype maps were built to study the structure of the categories according to the central core and peripheral areas. Cluster analysis was performed to corroborate the structure of the representations. Finally, multiple factor analysis (MFA) was performed to determine the similarity of the results obtained from the three samples using the RV coefficient. No statistical differences (p > 0.05) were identified in most of the representations (n = 11) generated from the WA task. Furthermore, the representation “Cheese and dairy products” was positioned in the central core of the three maps. The APs and the CSIs of each representation were similar in the three study samples. The RV coefficient (≥0.80) indicated similarity in the representations obtained. Results of this research can be useful for future studies that attempt to compare the stability and validity of the information based on qualitative and more flexible methodologies. Some methodological implications related to the validity and stability of projective methods are discussed.