農林漁牧普查爲台灣常川辦理之國勢普查之一，然因普查母體規模龐大，故每五年方能辦 理一次，爲翔實並即時反映每年農牧業之各項經營特徵，規畫針對農牧業經營概況之常規性年 度抽樣調查確有其必要性。本研究利用99年農林漁牧普查資料建構台灣主力農家所得調查的抽 樣策略，主力農家的定義爲戶內至少有一位65歲以下從農者,且初級農畜產品銷售額可達20萬 以上者。經考量我國執行農牧業經營概況調查之實際情形及各抽樣方法之利弊後，選定以最適 分層隨機抽樣設計作爲主力農家所得調查中之抽樣設計基礎，以農牧戶經營類別及規模作爲分 層變數，經最佳化分層演算法選定分層點，決定各層最佳化樣本數配置，並以模擬方式評估提出 之抽樣方法在估計精確度上的表現。

Conducted every five years, the census of Agriculture, Forestry, Fishery and Animal Husbandry is one of the important official surveys to help inform policy decisions. However, enormous survey cost and effort for the data processing are required for such a nationwide census. Certain annual agriculture surveys are necessary for the realization of current related industry information, so that proper policies can be formulated timely. The Major Farm Household Income Survey is an annual survey providing information on major farm household income. The main objective of this research is to construct a sampling strategy for the survey. A stratified random sampling design, in which the optimal stratum boundary and allocation are carried out based on the 2010 census data, is proposed for the purpose to enhance the estimation precision and investigate certain subpopulations of interest. The result indicates that the performance of the proposed stratified sampling is much more advantageous than simple random sampling without replacement and other stratified design with optimal allocation and equal within-stratum size stratification boundary under a comparable total sample size.