Latent Structure Analysis (LSA), particularly Latent Class Analysis (LCA), is introduced as a method to identify unobserved latent variables that explain the relationships between observed variables. It shares similarities with factor analysis but can also be viewed as a non-metric cluster analysis. The parameters of LCA model probabilities, and some estimation methods may produce invalid estimates. A sufficient sample size is necessary for meaningful application of maximum likelihood estimators, but very large samples can lead to significance problems in fit tests. A modified chi-square minimum estimation method by Mooijarta is suggested for its promising properties. Only marginal frequencies up to a certain order are considered in this method. A study is used to interpret the results of an LCA based on Mooijarta's method. New research directions for applying LCA in larger studies are also mentioned. LSA, as an exploratory method, has gained increasing interest in recent years. It is a very specific method, and more experience is needed in its application. LSA includes various methodologically interesting models, and most published studies focus on methodological research. The following discussion focuses on relatively simple models. The basic idea of LSA is to treat observed variables as indicators or symptoms of a latent property. The relationship between the latent variable and observed characteristics is assumed to be non-deterministic. In medical diagnosis, a patient may not necessarily have all typical symptoms of a disease, and some healthy patients may show symptoms. In such cases, a decision about the patient's health status must be made. The latent variable has two possible states, such as "sick" and "healthy." A latent structure model would allow the estimation of the probability that a sick person has certain symptoms or that the presence of certain symptoms indicates a specific illness. The probability that a sick person has certain symptoms corresponds to the parameters of latent structure models. When both latent and observed variables are non-metric, it is referred to as Latent Class Analysis (LCA). For the case of non-metric observed variables but metric latent variables, latent polynomial models are most important. Here, the probability that a person with a certain value on the latent continuum has certain manifestations on the observed variables is modeled as a polynomial. In latent profile analysis, the observed variables are metric.Latent Structure Analysis (LSA), particularly Latent Class Analysis (LCA), is introduced as a method to identify unobserved latent variables that explain the relationships between observed variables. It shares similarities with factor analysis but can also be viewed as a non-metric cluster analysis. The parameters of LCA model probabilities, and some estimation methods may produce invalid estimates. A sufficient sample size is necessary for meaningful application of maximum likelihood estimators, but very large samples can lead to significance problems in fit tests. A modified chi-square minimum estimation method by Mooijarta is suggested for its promising properties. Only marginal frequencies up to a certain order are considered in this method. A study is used to interpret the results of an LCA based on Mooijarta's method. New research directions for applying LCA in larger studies are also mentioned. LSA, as an exploratory method, has gained increasing interest in recent years. It is a very specific method, and more experience is needed in its application. LSA includes various methodologically interesting models, and most published studies focus on methodological research. The following discussion focuses on relatively simple models. The basic idea of LSA is to treat observed variables as indicators or symptoms of a latent property. The relationship between the latent variable and observed characteristics is assumed to be non-deterministic. In medical diagnosis, a patient may not necessarily have all typical symptoms of a disease, and some healthy patients may show symptoms. In such cases, a decision about the patient's health status must be made. The latent variable has two possible states, such as "sick" and "healthy." A latent structure model would allow the estimation of the probability that a sick person has certain symptoms or that the presence of certain symptoms indicates a specific illness. The probability that a sick person has certain symptoms corresponds to the parameters of latent structure models. When both latent and observed variables are non-metric, it is referred to as Latent Class Analysis (LCA). For the case of non-metric observed variables but metric latent variables, latent polynomial models are most important. Here, the probability that a person with a certain value on the latent continuum has certain manifestations on the observed variables is modeled as a polynomial. In latent profile analysis, the observed variables are metric.