diff --git a/README.md b/README.md
index ca982a0..e6ea2d2 100644
--- a/README.md
+++ b/README.md
@@ -85,6 +85,7 @@ Latte is still early stage software under intensive development.
 Start a Cassandra cluster somewhere (can be a local node). Then run:
 
 ```shell
+latte schema <workload.rn> [<node address>] # create the database schema 
 latte load <workload.rn> [<node address>]   # populate the database with data
 latte run <workload.rn> [<node address>]    # execute the workload and measure the performance 
  ```
@@ -124,13 +125,16 @@ Instance functions on `ctx` are asynchronous, so you should call `await` on them
 
 ### Schema creation
 
-You can create your own keyspaces and tables in the `schema` function:
+You can (re)create your own keyspaces and tables needed by the benchmark in the `schema` function.
+The `schema` function should also drop the old schema if present.
+The `schema` function is executed by running `latte schema` command.
 
 ```rust
 pub async fn schema(ctx) {
   ctx.execute("CREATE KEYSPACE IF NOT EXISTS test \
                  WITH REPLICATION = { 'class' : 'SimpleStrategy', 'replication_factor' : 1 }").await?;
-  ctx.execute("CREATE TABLE IF NOT EXISTS test.test(id bigint, data varchar").await?;
+  ctx.execute("DROP TABLE IF NOT EXISTS test.test").await?;
+  ctx.execute("CREATE TABLE test.test(id bigint, data varchar)").await?;
 }
 ```
 
diff --git a/src/config.rs b/src/config.rs
index 0b9735e..e8860e1 100644
--- a/src/config.rs
+++ b/src/config.rs
@@ -96,6 +96,26 @@ pub struct ConnectionConf {
     pub addresses: Vec<String>,
 }
 
+#[derive(Parser, Debug, Serialize, Deserialize)]
+#[clap(
+    setting(AppSettings::NextLineHelp),
+    setting(AppSettings::DeriveDisplayOrder)
+)]
+pub struct SchemaCommand {
+    /// Parameter values passed to the workload, accessible through param! macro.
+    #[clap(short('P'), parse(try_from_str = parse_key_val),
+    number_of_values = 1, multiple_occurrences = true)]
+    pub params: Vec<(String, String)>,
+
+    /// Path to the workload definition file.
+    #[clap(name = "workload", required = true, value_name = "PATH")]
+    pub workload: PathBuf,
+
+    // Cassandra connection settings.
+    #[clap(flatten)]
+    pub connection: ConnectionConf,
+}
+
 #[derive(Parser, Debug, Serialize, Deserialize)]
 #[clap(
     setting(AppSettings::NextLineHelp),
@@ -271,7 +291,17 @@ pub struct HdrCommand {
 #[derive(Parser, Debug)]
 #[allow(clippy::large_enum_variant)]
 pub enum Command {
-    /// Generates the data needed for the benchmark (typically needed by read benchmarks).
+    /// Creates the database schema by invoking the `schema` function of the workload script.
+    ///
+    /// The function should remove the old schema if present.
+    /// Calling this is likely to remove data from the database.
+    Schema(SchemaCommand),
+
+    /// Erases and generates fresh data needed for the benchmark by invoking the `erase` and `load`
+    /// functions of the workload script.
+    ///
+    /// Running this command is typically needed by read benchmarks.
+    /// You need to create the schema before.
     Load(LoadCommand),
 
     /// Runs the benchmark.
@@ -285,13 +315,11 @@ pub enum Command {
     /// Can compare two runs.
     Show(ShowCommand),
 
-    /// Exports call- and response-time histograms as a compressed HDR interval log.
+    /// Exports histograms as a compressed HDR interval log.
     ///
     /// To be used with HdrHistogram (https://github.com/HdrHistogram/HdrHistogram).
     /// Timestamps are given in seconds since Unix epoch.
     /// Response times are recorded in nanoseconds.
-    ///
-    /// Each histogram is tagged by the benchmark name, parameters and benchmark tags.
     Hdr(HdrCommand),
 }
 
diff --git a/src/main.rs b/src/main.rs
index 5c4a126..8358460 100644
--- a/src/main.rs
+++ b/src/main.rs
@@ -14,7 +14,8 @@ use tokio::runtime::{Builder, Runtime};
 use config::RunCommand;
 
 use crate::config::{
-    AppConfig, Command, ConnectionConf, HdrCommand, Interval, LoadCommand, ShowCommand,
+    AppConfig, Command, ConnectionConf, HdrCommand, Interval, LoadCommand, SchemaCommand,
+    ShowCommand,
 };
 use crate::context::*;
 use crate::context::{CassError, CassErrorKind, Context, SessionStats};
@@ -105,17 +106,29 @@ async fn connect(conf: &ConnectionConf) -> Result<(Context, Option<ClusterInfo>)
     Ok((session, cluster_info))
 }
 
-async fn load(conf: LoadCommand) -> Result<()> {
+/// Runs the `schema` function of the workload script.
+/// Exits with error if the `schema` function is not present or fails.
+async fn schema(conf: SchemaCommand) -> Result<()> {
     let mut program = load_workload_script(&conf.workload, &conf.params)?;
     let (mut session, _) = connect(&conf.connection).await?;
-
-    if program.has_schema() {
-        eprintln!("info: Creating schema...");
-        if let Err(e) = program.schema(&mut session).await {
-            eprintln!("error: Failed to create schema: {}", e);
-            exit(255);
-        }
+    if !program.has_schema() {
+        eprintln!("error: Function `schema` not found in the workload script.");
+        exit(255);
     }
+    eprintln!("info: Creating schema...");
+    if let Err(e) = program.schema(&mut session).await {
+        eprintln!("error: Failed to create schema: {}", e);
+        exit(255);
+    }
+    eprintln!("info: Schema created successfully");
+    Ok(())
+}
+
+/// Loads the data into the database.
+/// Exits with error if the `load` function is not present or fails.
+async fn load(conf: LoadCommand) -> Result<()> {
+    let mut program = load_workload_script(&conf.workload, &conf.params)?;
+    let (mut session, _) = connect(&conf.connection).await?;
 
     if program.has_prepare() {
         eprintln!("info: Preparing...");
@@ -184,14 +197,6 @@ async fn run(conf: RunCommand) -> Result<()> {
         conf.cass_version = Some(cluster_info.cassandra_version);
     }
 
-    if program.has_schema() {
-        eprintln!("info: Creating schema...");
-        if let Err(e) = program.schema(&mut session).await {
-            eprintln!("error: Failed to create schema: {}", e);
-            exit(255);
-        }
-    }
-
     if program.has_prepare() {
         eprintln!("info: Preparing...");
         if let Err(e) = program.prepare(&mut session).await {
@@ -347,6 +352,7 @@ async fn export_hdr_log(conf: HdrCommand) -> Result<()> {
 
 async fn async_main(command: Command) -> Result<()> {
     match command {
+        Command::Schema(config) => schema(config).await?,
         Command::Load(config) => load(config).await?,
         Command::Run(config) => run(config).await?,
         Command::Show(config) => show(config).await?,
diff --git a/workloads/basic/read.rn b/workloads/basic/read.rn
index c8d8147..5143e69 100644
--- a/workloads/basic/read.rn
+++ b/workloads/basic/read.rn
@@ -11,7 +11,8 @@ const TABLE = "basic";
 pub async fn schema(ctx) {
     ctx.execute(`CREATE KEYSPACE IF NOT EXISTS ${KEYSPACE} \
                     WITH REPLICATION = { 'class' : 'SimpleStrategy', 'replication_factor' : 1 }`).await?;
-    ctx.execute(`CREATE TABLE IF NOT EXISTS ${KEYSPACE}.${TABLE}(id bigint PRIMARY KEY)`).await?;
+    ctx.execute(`DROP TABLE IF EXISTS ${KEYSPACE}.${TABLE}`).await?;
+    ctx.execute(`CREATE TABLE ${KEYSPACE}.${TABLE}(id bigint PRIMARY KEY)`).await?;
 }
 
 pub async fn erase(ctx) {
diff --git a/workloads/basic/write-blob.rn b/workloads/basic/write-blob.rn
index dc47666..9e7b748 100644
--- a/workloads/basic/write-blob.rn
+++ b/workloads/basic/write-blob.rn
@@ -8,7 +8,8 @@ const TABLE = "blob";
 pub async fn schema(db) {
     db.execute(`CREATE KEYSPACE IF NOT EXISTS ${KEYSPACE} \
                     WITH REPLICATION = { 'class' : 'SimpleStrategy', 'replication_factor' : 1 }`).await?;
-    db.execute(`CREATE TABLE IF NOT EXISTS ${KEYSPACE}.${TABLE}(id bigint PRIMARY KEY, data BLOB)`).await?;
+    db.execute(`DROP TABLE IF EXISTS ${KEYSPACE}.${TABLE}`).await?;
+    db.execute(`CREATE TABLE ${KEYSPACE}.${TABLE}(id bigint PRIMARY KEY, data BLOB)`).await?;
 }
 
 pub async fn erase(db) {
diff --git a/workloads/basic/write.rn b/workloads/basic/write.rn
index 84ab967..e307a9b 100644
--- a/workloads/basic/write.rn
+++ b/workloads/basic/write.rn
@@ -8,7 +8,8 @@ const TABLE = "basic";
 pub async fn schema(db) {
     db.execute(`CREATE KEYSPACE IF NOT EXISTS ${KEYSPACE} \
                     WITH REPLICATION = { 'class' : 'SimpleStrategy', 'replication_factor' : 1 }`).await?;
-    db.execute(`CREATE TABLE IF NOT EXISTS ${KEYSPACE}.${TABLE}(id bigint PRIMARY KEY)`).await?;
+    db.execute(`DROP TABLE IF EXISTS ${KEYSPACE}.${TABLE}`).await?;
+    db.execute(`CREATE TABLE ${KEYSPACE}.${TABLE}(id bigint PRIMARY KEY)`).await?;
 }
 
 pub async fn erase(db) {
diff --git a/workloads/sai/new/common.rn b/workloads/sai/new/common.rn
index 56f3e68..018c0fd 100644
--- a/workloads/sai/new/common.rn
+++ b/workloads/sai/new/common.rn
@@ -26,7 +26,9 @@ pub async fn init_schema(db) {
         CREATE KEYSPACE IF NOT EXISTS ${KEYSPACE}
             WITH replication = {'class': 'SimpleStrategy', 'replication_factor': 1 }`).await?;
     db.execute(`
-        CREATE TABLE IF NOT EXISTS ${KEYSPACE}.${TABLE} (
+        DROP TABLE IF EXISTS ${KEYSPACE}.${TABLE}`).await?;
+    db.execute(`
+        CREATE TABLE ${KEYSPACE}.${TABLE} (
             par_id bigint,
             row_id uuid,
             time1 timestamp,
diff --git a/workloads/sai/orig/lib.rn b/workloads/sai/orig/lib.rn
index 375dc84..ae8bbaf 100644
--- a/workloads/sai/orig/lib.rn
+++ b/workloads/sai/orig/lib.rn
@@ -20,7 +20,9 @@ pub async fn schema(ctx) {
         CREATE KEYSPACE IF NOT EXISTS ${KEYSPACE}
             WITH replication = {'class': 'SimpleStrategy', 'replication_factor': 1 }`).await?;
     ctx.execute(`
-        CREATE TABLE IF NOT EXISTS ${KEYSPACE}.${TABLE} (
+        DROP TABLE IF EXISTS ${KEYSPACE}.${TABLE}`).await?;
+    ctx.execute(`
+        CREATE TABLE ${KEYSPACE}.${TABLE} (
             id uuid,
             time timestamp,
             value int,