Nowadays, the business is highly competitive. In online channels, it can be considered as another way to reach customers. Customers who are unable to live a normal life turn to online channels to make purchases. In the online channels, finding customer behavior is one of the still problems. It is hard to know the actual needs of real customers through the online purchasing process. Therefore, this research uses the concept of past purchasing behavior. The principle is that customers with the same purchasing behavior will have the same purchasing behavior. All purchased transaction were classified to the group by using the similarity of purchasing. The purchased transactions of the new customer were calculated with each group buy using Euclidean distance. The shortest distance between the purchased transactions of the new customer and ith group is defined as the target group. The information in the target group is used to recommend products to the customer. The reason that why the recommendation has very importance is there are many amount products available. As a result, it is difficult for customers to access the products they want. This research presents recommendation system by using customer segmentation. The k-Mean was used to classify the customer segmentation. The data in each group is analyzed for frequency of purchases. Which products have a higher purchase frequency are more significant. Therefore, the recommendation system uses the top-N high-frequency products to recommend to the active customer. To evaluate the recommendation system, the F-Measure method is used to measure the efficiency system. The F-Measure consists of F1, Precision, and Recall. F1 is the harmonic mean. The dataset collected from the online marketing FBT Life Limited Partnership of 287 people, 41 items, 1,147 number of transactions. The dataset is divided into training set and test set. The 10-Fold cross validation technique is to evaluate the effectiveness of the recommendation system. In the k-Mean, the number of clusters is defined as k = 3, 5, 7, 9, and 11. To test the Top-N recommendation system, N is the number of products for recommending for the active customer. The N is defined as 5 and 10. The experimental results showed that k = 7 gave the best recommendation efficiency on Top-5 with F1 value of 31.11%, Precision value of 28%, and Recall value of 35%. In Top-10, the recommendation efficiency F1, Precision, and Recall are 31.97%, 31%, and 33%, respectively.