diff --git a/petab/v2/normalize.py b/petab/v2/normalize.py
new file mode 100644
index 00000000..a2258a29
--- /dev/null
+++ b/petab/v2/normalize.py
@@ -0,0 +1,160 @@
+"""Functions for normalizing PEtab tables.
+
+Setting data types, adding missing columns and replacing NA by default values.
+"""
+
+from typing import TypeVar
+
+import pandas as pd
+
+from .C import *
+
+__all__ = [
+    "normalize_parameter_df",
+    "normalize_measurement_df",
+    "normalize_condition_df",
+    "normalize_observable_df",
+]
+
+DataFrameOrNone = TypeVar("DataFrameOrNone", pd.DataFrame, None)
+
+
+def normalize_parameter_df(
+    df: DataFrameOrNone, inplace: bool = False
+) -> DataFrameOrNone:
+    """Normalize parameter table.
+
+    Arguments:
+        df: Parameter table
+        inplace: Modify DataFrame in place
+    Returns:
+        The updated DataFrame
+    """
+    if df is None:
+        return
+
+    if not inplace:
+        df = df.copy()
+
+    col_to_type = {
+        PARAMETER_ID: str,
+        PARAMETER_NAME: str,
+        PARAMETER_SCALE: str,
+        NOMINAL_VALUE: float,
+        LOWER_BOUND: float,
+        UPPER_BOUND: float,
+        ESTIMATE: bool,
+        INITIALIZATION_PRIOR_TYPE: str,
+        INITIALIZATION_PRIOR_PARAMETERS: str,  # TODO -> tuple?
+        OBJECTIVE_PRIOR_TYPE: str,
+        OBJECTIVE_PRIOR_PARAMETERS: str,  # TODO -> tuple?
+    }
+    for col, dtype in col_to_type.items():
+        if col in df.columns:
+            df[col] = df[col].astype(dtype)
+        else:
+            df[col] = pd.NA
+
+    # TODO only if estimate is True?
+    df[INITIALIZATION_PRIOR_TYPE] = df[INITIALIZATION_PRIOR_TYPE].fillna(
+        PARAMETER_SCALE_UNIFORM
+    )
+    # TODO: lb;ub
+    #  df[INITIALIZATION_PRIOR_PARAMETERS]
+
+    return df
+
+
+def normalize_measurement_df(
+    df: DataFrameOrNone, inplace: bool = False
+) -> DataFrameOrNone:
+    """Normalize measurement table.
+
+    Arguments:
+        df: Measurement table
+        inplace: Modify DataFrame in place
+    Returns:
+        The updated DataFrame
+    """
+    if df is None:
+        return
+
+    if not inplace:
+        df = df.copy()
+
+    col_to_type = {
+        OBSERVABLE_ID: str,
+        PREEQUILIBRATION_CONDITION_ID: str,
+        SIMULATION_CONDITION_ID: str,
+        MEASUREMENT: float,
+        TIME: float,
+        OBSERVABLE_PARAMETERS: str,  # TODO -> tuple?
+        NOISE_PARAMETERS: str,  # TODO -> tuple?
+    }
+    for col, dtype in col_to_type.items():
+        if col in df.columns:
+            df[col] = df[col].astype(dtype)
+        else:
+            df[col] = pd.NA
+
+    return df
+
+
+def normalize_condition_df(
+    df: DataFrameOrNone, inplace: bool = False
+) -> DataFrameOrNone:
+    """Normalize condition table.
+
+    Arguments:
+        df: Condition table
+        inplace: Modify DataFrame in place
+    Returns:
+        The updated DataFrame
+    """
+    if df is None:
+        return
+
+    if not inplace:
+        df = df.copy()
+
+    # TODO: always as string even if everything is numeric?
+    # TODO: everything but ID and Name to sympy?
+    df = df.astype(str)
+
+    return df
+
+
+def normalize_observable_df(
+    df: DataFrameOrNone, inplace: bool = False
+) -> DataFrameOrNone:
+    """Normalize observable table.
+
+    Arguments:
+        df: Observable table
+        inplace: Modify DataFrame in place
+    Returns:
+        The updated DataFrame
+    """
+    if df is None:
+        return
+
+    if not inplace:
+        df = df.copy()
+
+    col_to_type = {
+        OBSERVABLE_ID: str,
+        OBSERVABLE_NAME: str,
+        OBSERVABLE_FORMULA: str,  # TODO -> sympy?
+        NOISE_FORMULA: str,  # TODO -> sympy?
+        NOISE_DISTRIBUTION: str,
+        OBSERVABLE_TRANSFORMATION: str,
+    }
+    for col, dtype in col_to_type.items():
+        if col in df.columns:
+            df[col] = df[col].astype(dtype)
+        else:
+            df[col] = pd.NA
+
+    df[OBSERVABLE_TRANSFORMATION] = df[OBSERVABLE_TRANSFORMATION].fillna(LIN)
+    df[NOISE_DISTRIBUTION] = df[NOISE_DISTRIBUTION].fillna(NORMAL)
+    df[NOISE_FORMULA] = df[NOISE_FORMULA].fillna(1.0)
diff --git a/petab/v2/problem.py b/petab/v2/problem.py
index 612f2571..586812da 100644
--- a/petab/v2/problem.py
+++ b/petab/v2/problem.py
@@ -24,6 +24,7 @@
 from ..v1.C import *  # noqa: F403
 from ..v1.models.model import Model, model_factory
 from ..v1.yaml import get_path_prefix
+from .normalize import *
 
 if TYPE_CHECKING:
     from ..v2.lint import ValidationIssue, ValidationResultList, ValidationTask
@@ -717,3 +718,22 @@ def validate(
                     break
 
         return validation_results
+
+    def normalize(self):
+        """Normalize tables.
+
+        This function normalizes the tables in the PEtab problem.
+        It sets data types, adds missing columns, and replaces NA by default
+        values.
+        """
+        normalize_condition_df(self.condition_df, inplace=True)
+        normalize_measurement_df(self.measurement_df, inplace=True)
+        normalize_parameter_df(self.parameter_df, inplace=True)
+        normalize_observable_df(self.observable_df, inplace=True)
+        # TODO
+        # normalize_mapping_df(self.mapping_df, inplace=True)
+
+        if self.visualization_df is not None:
+            from petab.v1.visualize.lint import _apply_defaults
+
+            _apply_defaults(self.visualization_df)