In the manufacturing industry, it is extremely important to identify variables that affect product quality. Identifying variables which affect quality variables is called causal analysis. In batch processes, time-series data of process variables and the corresponding data of quality variables are generally acquired. Since causal analysis using the raw data needs a large computation load, it is often performed after compressing time-series process variables data into non-time-series feature variables data. Various causal analysis methods using such data have been developed, however, none have shown effective results in actual plants. In the present work, non-time-series kernel Granger causality (NTS-KGC) is proposed for causal analysis with non-time-series data of batch processes. This is a method that kernel Granger causality [1], which is used for causal analysis with time-series data in nonlinear systems, is expanded for causal analysis with non-time-series data. The validity of the proposed method is demonstrated through a numerical example of a nonlinear batch process. In addition, we conducted a case study of applying NTS-KGC to data obtained from a real steelmaking process. The results demonstrate that NTS-KGC is superior to other existing methods using the following indexes, i.e. variable influence on projection (VIP) of partial least squares (PLS), regression coefficients of PLS, and variable importance of Random Forest.