The difficulty of a survey question is known to be a serious source of response error (Bless et al., 1992; Knäuper et al., 1997; Krosnick, 1991). If questions are difficult to understand respondents are likely to give incorrect answers, they may become frustrated, they may guess, or they may even refuse to answer any further question. An important goal in questionnaire design is to write survey questions that are easy to understand and to answer. However, very little is known about the factors that determine question difficulty. Only recently have survey researchers started to look at specific text features in order to explain why some questions are easier to comprehend or have less cognitive burden than others (Lessler&Forsyth, 1996; Graesser et al., 2006; Tourangeau, 2000). Theoretical and empirical evidence from psycholinguistics suggests that these features (e.g. low-frequency words, vague nounphrases, left-embedded syntax) cause comprehension difficulties and can thus have a strong impact on survey response quality. In order to examine the effects of these text features, we conducted an online experiment in which two versions of the same question were compared using response latency times (Draisma&Dijkstra, 2004). One group (N=45) received well-formulated survey questions, the other group (N=47) answered questions which were suboptimal with respect to several psycholinguistic text features. The results show that the text features influenced the cognitive effort required to answer survey questions. This was tested with an ANCOVA (with reading rate as a covariate) using overall response latency time as the dependant variable. It revealed a significant difference between both conditions (p=0.02) with subjects answering the suboptimal questions having longer overall response latency times. Moreover, question difficulty affected the distribution of answers Hence, questionnaire designers are advised to pay attention to these text features when crafting questions. The experiment is currently replicated with a larger sample and a slightly modified questionnaire. Besides response latencies and answer distributions, also dropout will be analyzed and presented.