During the last three decades, a variety of cointegration tests has been developed to investigate the long-run relationship between two data series. All these tests are based on different characteristics/structures of original time series. These tests were assessed for their performance on basis of their size and power properties using Monte Carlo simulations in a number of studies like, Banerjee, Dolado et al. (1986), Kremers, Ericsson et al. (1992), Haug (1996), Mariel (1996), Österholm (2003), Gabriel (2003), Pesavento (2004) and many more. However, these studies did not give a decisive conclusion that which test is performing better than rest. This is due to variant features of underlying series and every test has its domain of weaknesses and strengths. As it is the basic principle of comparison that tests can be compared on basis of their power properties only when all of tests have same size or stable size around nominal size. However, all of these comparisons were carried out using asymptotic critical values and their sizes were not same and stable. As almost all of comparisons were based on Monte Carlo simulations and the design of Monte Carlo supports the assumptions of model and in real data one doesn’t know whether the assumptions of model hold or not, so there was a gap in literature that these tests may be compared on basis of their size and power properties in a universal set of alternatives using Monte Carlo methods and also these tests may be compared on basis of some real economic data. To fill this gap, this study was carried out in which 24 cointegration tests were assessed for their performance. Out of these 24 tests, 16 tests belonged to the class with null hypothesis of no cointegration and 8 tests belonged to the class with null hypothesis ii of cointegration. These tests were evaluated on basis of stringency criterion (Zaman 1996) using Monte Carlo methods and on basis of Empirical Size and Power properties using real data of income and consumption of 100 countries from 1970 to 2014. First tests were assessed for their size stability using asymptotic critical values and then size of these tests was stabilized using simulated critical values. All of 24 tests were assessed on basis of stringency criterion using simulated critical values by considering two assumptions, one that the nature of deterministic part was known a priori and second that the nature of deterministic part was unknown, so an automatic selection procedure of deterministic part (Elder and Kennedy 2001) was followed. As it is strongly recommended by economic theory that income and consumption of same country are cointegrated and income and consumption of two different countries are not cointegrated, so on basis of this economic theory all of 24 tests were evaluated for their performance on basis of Empirical Size and Power. The study revealed that almost all of tests i.e. 21 out 24 had unstable size when asymptotic critical values were used and all of tests had stable size when simulated critical values were used. From tests with null hypothesis of no cointegration, Phillips and Ouliaris ˆ P m test was the leading performer as it was the most stringent test at all specifications of deterministic part. In addition to this test Choi Durbin-Hausman test, Phillips and Ouliaris’ ˆt Z test, Phillips and Ouliaris’ ˆ Z a test and the t-test of cointegration in a single equation Error Correction Model also performed overall better than rest of tests. Three tests i.e. Johansen Trace, Johansen Maximum Eigen Value and Phillips and Ouliaris ˆz P performed overall poor. In same manner from tests with null hypothesis of cointegration LM test based on KPSS statistic was the leading performer as it was the iii most stringent test for all of three specifications of deterministic part. Shin’s C test was the second overall better performing test and three tests i.e. Xiao Fluctuation, Hansen’s cL and Hausman H1 overall performed poorly. It was also revealed that on basis of real data comparison tests with null hypothesis of no cointegration performed way well as compared to tests with null hypothesis of cointegration. On basis of real data, from tests with null hypothesis of no cointegration five tests i.e. the t-test of cointegration in a single equation Error Correction Model, Boswijk Wald test, Phillips and Ouliaris ˆ P m test, Phillips and Ouliaris’ ˆ Z a test and Choi Durbin-Hausman test were the leading performers both in terms of Empirical Size and Power. However, on basis of real data, from test with null hypothesis of cointegration not a single test had performance that is worth to mention. It is recommended in light of our study that use of asymptotic critical values may be eluded and instead simulated critical values may be used. It is also recommended that tests with null hypothesis of no cointegration may be given preference over tests with null hypothesis of cointegration for a specific real economic problem and if tests with null hypothesis of cointegration are used then, they may only be used for confirmatory purposes. Five tests with null hypothesis of no cointegration are recommended and these are Phillips and Ouliaris ˆ P m test, the t-test of cointegration in a single equation Error Correction Model, Boswijk Wald test, Phillips and Ouliaris’ ˆ Z a test and Choi Durbin-Hausman test. Similarly, from tests with null hypothesis of cointegration only one test i.e. LM test based on KPSS statistic may be used for confirmatory purpose.