The paper focuses on one of the main longitudinal surveys conducted by the US Bureau of the Census, the Survey of Income and Program Participation (SIPP); specifically it regards the problem of estimating gross flows in the labour market when data are affected by classification errors. Some data analysis presented in the paper show that typical sources of errors in longitudinal data, like time-in-sample effect, do not affect SIPP observed gross flows dramatically. The so-called seam effect dominates on other phenomena that potentially bias gross flows observed by means of longitudinal surveys. On the basis of this evidence, a strategy to correct labour force flows observed with SIPP from classification errors is proposed. The strategy is an ad hoc specification of latent class analysis. Gross flows among labour force states are a major tool to analyse the dynamics of labour market. In most industrialised countries the central statistical office conducts longitudinal surveys in order to collect information on labour force participation. Longitudinal surveys follow units over time; they may be distinguished between (i) panel surveys, where information from the same respondents is collected at successive moments of time and (ii) retrospective surveys, cross-sectional surveys including retrospective questions; more often, in practical situations, they are a mixture of the two typologies. To analyse labour market dynamics, usually individuals are classified into three mutually exclusive states: Employed (E), Unemployed (U) and Not in the labour force (N). With this information net flows and gross flows may be estimated. Gross flows regard changes at individual level. Longitudinal data, whether repeated or retrospective, are always affected by measurement errors. With reference to gross flows estimation in the labour market, these errors are termed classification errors, since they cause individuals to be erroneously classified with regard to their position in the labour market. Even if sometimes these errors may cancel out when estimating net flows, they can not be ignored when estimating gross flows. Sources of classification errors for data collected by panel surveys are reviewed in [5]. Panel data are affected by the so-called panel attrition, which regards essentially unit nonresponse, either because individuals leave the sample before the end of