The deterministic seismic inversion method has been successfully used in various projects in exploration and development. This method enables the interpreter to get a better understanding of subsurface by omitting the wavelet and tuning effects; therefore, quantitative reservoir properties can be generated. However, this method has significant limitations by generating average impedances of each layer, and the range of values is smaller than the impedance from the wells; therefore, the inversion will not produce results that are not within the calibration range. Stochastic seismic inversion is done by conditioning well data and reproducing spatially varying statistics using a variogram, which could overcome the deterministic limitation. This method generates multiple realizations of high-frequency elastic properties that are consistent with both seismic amplitude and well data. These two methods are applied in "K" gas field, which located in the offshore Bonaparte Basin, Eastern Indonesia. Geologically, the "K" field is located within a relatively undeformed Australian continental margin that extends into Indonesian waters. This field contains a significant gas column, reservoir within shallow marine, highly mature, quartzose sandstone of the Middle Jurassic Plover Formation. Potential targets in the area may be large folds, horst blocks, and tilted fault blocks trap in the Palaeozoic section. The application of stochastic seismic inversion showed significant benefits compared to deterministic, especially in "K" gas field, where the reservoirs are stacked sandstone with intra-formational shale. Some of the reservoir and all the intraformational shales are below seismic resolution. Stochastic seismic inversion is able to distinguish those features, and in addition, the inverted volumes with multiple realizations with ranking criteria for P10, P50, and P90 of impedance, and vp/vs could be utilized to reduce the risk associated with exploration plan and field development. In such instances, the stochastic seismic inversion method could provide the uncertainties associated with the models that have been generated.