This study was conducted to determine the causes and location of student difficulties in solving mathematical problems with HOTS based on the mathematical connection skills they have. This research using qualitative methods, and was conducted in the even semester of the 2017/2018 academic year at STKIP Campus Santu Paulus Ruteng. The research subjects were selected using purposive sampling technique. The selected items were three students of the STKIP Santu Paulus Ruteng Mathematics Education Study Program, with categories namely one student capable of deep mathematical connections, one student capable of moderate mathematical connections, and one student capable of high mathematical connections. The selection of subjects from each of these categories is based on the results of the identification of problems that have been carried out during the initial observation. The instrument in this study was the researchers themselves. Besides, researchers use supporting instruments, namely tests and guidelines for interviews. In conducting interviews, researchers conduct structured and unstructured interviews. The technique of data collection is done by giving written tests and interviews conducted through the triangulation process. The triangulation process aims to obtain valid data carried out by conducting repeated interviews at different times to the same research subject. The interval between the first phase of the interview with the next stage is 14 days, and because the data is saturated, so the data collection stops. Data analysis is carried out when data collection takes place, and after data collection is carried out. The steps in analyzing the data using the Miles and Huberman analysis models, namely 1) reducing data, 2) presenting data, and 3) concluding. The results showed that, in solving mathematical problems with Hots, the subject of KKMR only relied on factual knowledge but had difficulties in using conceptual, procedural and metalogical knowledge, while the KKMS subject could only use factual and procedural knowledge but had difficulty using procedural knowledge and metacognition, while KKMT subjects only had difficulty in using their metalogical knowledge but did not experience difficulties in using factual, conceptual and procedural knowledge. Learning difficulties experienced by each subject are caused because they do not often practice mathematical solving with Hots when learning mathematics.