Gold is an important financial asset due to its special characteristics. The increasing public interest in gold investment drives the importance of gold price prediction in making investment decisions. Gold price prediction is challenging because it is influenced by several economic factors such as the rate of inflation, stock prices, exchange rates, and value of other commodities. This research aims to determine an accurate model in predicting gold prices in Indonesia using the Long Short-Term Memory (LSTM) method. Two types of LSTM models are compared: multivariate LSTM and univariate LSTM. In multivariate LSTM, eight economic variables are used as predictors: inflation rate, BI rate, crude price, S&P stock price, DJI stock price, IHSG stock price, and exchange rate (IDR/USD). The accuracy of the testing dataset with all economic variables gave a result of 98.4%. The feature selection with wrapper method shows there are three variables that have a significant influence: BI Rate, IHSG, and inflation rate. The accuracy of the testing dataset with these three economic variables is 97.4%. Univariate time series LSTM provides better accuracy, which is 98.9%. Compared to multivariate LSTM, univariate LSTM is superior because of higher accuracy, and ease in collecting the required data.